cogito/calculet-npu-research-archive
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
main
calculet-npu-research-archive/history/llama.cpp/llama.cpp-commits-all.tsv
T

1.7 MiB
Raw Permalink Blame History

1fd9bd632f346085920d523dd307a3e4c11cc0a05e08e473cf292ba73789769b993bb35bcdfc31689wangbomengwangbomeng@calculet.tech2026-06-01T14:15:35+08:00wangbomengwangbomeng@calculet.tech2026-06-01T14:15:35+08:00HEAD -> dev, origin/devllama-bench: fix device-info
2e08e473cf292ba73789769b993bb35bcdfc31689f86112d6758298241ab2fa84ad4f0f7b1ace35c6wangbomengwangbomeng@calculet.tech2026-06-01T11:47:01+08:00wangbomengwangbomeng@calculet.tech2026-06-01T11:47:01+08:00llama-bench:add device-info
3f86112d6758298241ab2fa84ad4f0f7b1ace35c6d1bc743c10080fed1253686ce01749b42611ded4wangbomengwangbomeng@calculet.tech2026-05-29T16:40:38+08:00wangbomengwangbomeng@calculet.tech2026-05-29T16:40:38+08:00add dump_pos,fix pos_tensor of pre_by_embeds
4d1bc743c10080fed1253686ce01749b42611ded42f201160b1960fd7be18b70d5770599d7082dd2fwangbomengwangbomeng@calculet.tech2026-05-28T14:23:06+08:00wangbomengwangbomeng@calculet.tech2026-05-28T14:23:06+08:00set u_batch = max_seq_len
52f201160b1960fd7be18b70d5770599d7082dd2ff131bf1908b8c5dd1f49d7ae7f616506fc471666wangbomengwangbomeng@calculet.tech2026-05-27T09:58:31+08:00wangbomengwangbomeng@calculet.tech2026-05-27T09:58:31+08:00delete useless code
6f131bf1908b8c5dd1f49d7ae7f616506fc47166633d96dae28e17208ef61a71dba40cda3c04f135awangbomengwangbomeng@calculet.tech2026-05-27T09:38:51+08:00wangbomengwangbomeng@calculet.tech2026-05-27T09:38:51+08:00fix patch of qwen2.5vl,n_patch of minicpm = 64,fix n_ctx_slot,comment out M-ROPE
7ebce9e1a548c2329aa97f53b8b2ad50d968088dce850ff4de25ad1221abc9bf3bb30df252ccd9c36Yunzhe Jiayunzhe@calculet.tech2026-05-27T09:38:00+08:00Yunzhe Jiayunzhe@calculet.tech2026-05-27T09:38:00+08:00origin/runtime_replaceRefactor cal-llm integration by removing runtime capacity loading and adjusting request handling
8e850ff4de25ad1221abc9bf3bb30df252ccd9c369985688f7fa96f755882f96a2cd8a18114fdc42dYunzhe Jiayunzhe@calculet.tech2026-05-26T10:59:45+08:00Yunzhe Jiayunzhe@calculet.tech2026-05-26T10:59:45+08:00Enhance cal-llm integration by adding runtime checks for CALRT_LIBRARY and updating link settings for Linux and Windows
99985688f7fa96f755882f96a2cd8a18114fdc42d38ca6895815671a87ceec86b4aa4eb976580ce76Yunzhe Jiayunzhe@calculet.tech2026-05-25T09:05:49+08:00Yunzhe Jiayunzhe@calculet.tech2026-05-25T09:05:49+08:00Refactor CalrtEngineConfig initialization by removing redundant parameters
1033d96dae28e17208ef61a71dba40cda3c04f135a37f068d87cc2d0a0b40ce978031fc9932fc89077wangbomengwangbomeng@calculet.tech2026-05-22T14:33:45+08:00wangbomengwangbomeng@calculet.tech2026-05-22T14:33:45+08:00add llama-build.sh
111921024604db640f09ade83c8a437ebf15bde360d028722a12c5a463cc97207787cfd03d7a2590b0 5af229fe518ee2d761ab8ae2a69f7a9aaa9057f4Bomeng Wangwangbomeng@calculet.tech2026-05-21T14:08:14+08:00Bomeng Wangwangbomeng@calculet.tech2026-05-21T14:08:14+08:00refs/stashWIP on dev: d028722a replace cparam.n_ubatch with n_ubatch()
125af229fe518ee2d761ab8ae2a69f7a9aaa9057f4d028722a12c5a463cc97207787cfd03d7a2590b0Bomeng Wangwangbomeng@calculet.tech2026-05-21T14:08:14+08:00Bomeng Wangwangbomeng@calculet.tech2026-05-21T14:08:14+08:00index on dev: d028722a replace cparam.n_ubatch with n_ubatch()
1337f068d87cc2d0a0b40ce978031fc9932fc8907713508e5e18bbb1202ae60c2ee5ecbdda1cca817bwangbomengwangbomeng@calculet.tech2026-05-21T14:07:00+08:00wangbomengwangbomeng@calculet.tech2026-05-21T14:07:00+08:00release
1413508e5e18bbb1202ae60c2ee5ecbdda1cca817bb6b7919468126d5ba30a10941825154789e9082fwangbomengwangbomeng@calculet.tech2026-05-21T11:43:59+08:00wangbomengwangbomeng@calculet.tech2026-05-21T11:43:59+08:00reduce cmake log
15b6b7919468126d5ba30a10941825154789e9082f46847c24fc5c52dd57bb4fb17fc4d684149bb4fawangbomengwangbomeng@calculet.tech2026-05-20T16:40:33+08:00wangbomengwangbomeng@calculet.tech2026-05-20T16:40:33+08:00reduce cmake log
1638ca6895815671a87ceec86b4aa4eb976580ce76ddd60c928224b52eb4e1e60e01e8a1ecbb80ee45Yunzhe Jiayunzhe@calculet.tech2026-05-20T16:37:13+08:00Yunzhe Jiayunzhe@calculet.tech2026-05-20T16:37:13+08:00Add support for cal-llm profile dumping to JSONL file
1746847c24fc5c52dd57bb4fb17fc4d684149bb4fad028722a12c5a463cc97207787cfd03d7a2590b0wangbomengwangbomeng@calculet.tech2026-05-20T16:02:12+08:00wangbomengwangbomeng@calculet.tech2026-05-20T16:02:12+08:00Simplify CMakeLists.txt
18d028722a12c5a463cc97207787cfd03d7a2590b063442fde8dc285817f725bd6c5fae80f478a3813wangbomengwangbomeng@calculet.tech2026-05-20T14:41:05+08:00wangbomengwangbomeng@calculet.tech2026-05-20T14:41:05+08:00replace cparam.n_ubatch with n_ubatch()
1963442fde8dc285817f725bd6c5fae80f478a3813fc3f4a9fab88e98eff8eeca8cbc6617333edba2cwangbomengwangbomeng@calculet.tech2026-05-20T11:47:42+08:00wangbomengwangbomeng@calculet.tech2026-05-20T11:47:42+08:00fix ubatch and cmakelist
20ddd60c928224b52eb4e1e60e01e8a1ecbb80ee452716130010c58f1cdd9d94f0e2f8e047bc92dc66Yunzhe Jiayunzhe@calculet.tech2026-05-20T09:22:40+08:00Yunzhe Jiayunzhe@calculet.tech2026-05-20T09:27:06+08:00Add cal-llm backend profiling support with command-line option
21fc3f4a9fab88e98eff8eeca8cbc6617333edba2cc931ef3163940227c9b6291e04216245cce21429wangbomengwangbomeng@calculet.tech2026-05-19T16:54:16+08:00wangbomengwangbomeng@calculet.tech2026-05-19T16:54:16+08:00add dump and rt case generation
22c931ef3163940227c9b6291e04216245cce21429a43741244a646d3f8fa3ff01bb9b7d059223d0a5wangbomengwangbomeng@calculet.tech2026-05-18T17:34:37+08:00wangbomengwangbomeng@calculet.tech2026-05-18T17:34:37+08:00calrt: fix prefill_by_ids.fix create_buf
232716130010c58f1cdd9d94f0e2f8e047bc92dc66b35bda124ccd4ed581dd6088d3ffa4931c2809b6Yunzhe Jiayunzhe@calculet.tech2026-05-18T08:40:54+08:00Yunzhe Jiayunzhe@calculet.tech2026-05-18T08:40:54+08:00fix n_ctx_slot error by adding runtime capacity loading for cal-llm
24a43741244a646d3f8fa3ff01bb9b7d059223d0a5b2003d6c90ef0fa26ab9e930ba3e22b355eb3d39wangbomengwangbomeng@calculet.tech2026-05-15T14:39:53+08:00wangbomengwangbomeng@calculet.tech2026-05-15T14:39:53+08:00run llama-passkey successfully, fix llama-bench, copy 3 times input to infer qwen2.5vl
25b35bda124ccd4ed581dd6088d3ffa4931c2809b681a2cbd4cef5d28334ff513d6b69e593d9d1cb49Yunzhe Jiayunzhe@calculet.tech2026-05-15T09:19:31+08:00Yunzhe Jiayunzhe@calculet.tech2026-05-15T09:19:31+08:00Implement vocab-only model initialization and enhance cal-llm backend error handling
2681a2cbd4cef5d28334ff513d6b69e593d9d1cb493b14ed556c15529fe1f94b649f49db7156dfaf67Yunzhe Jiayunzhe@calculet.tech2026-05-14T16:06:37+08:00Yunzhe Jiayunzhe@calculet.tech2026-05-14T16:06:37+08:00Enhance CMake configuration for cal-llm integration by updating paths and adding library properties
273b14ed556c15529fe1f94b649f49db7156dfaf67dde8b8228081769c8d8f1e0751d47fa7875f8423Yunzhe Jiayunzhe@calculet.tech2026-05-14T14:34:25+08:00Yunzhe Jiayunzhe@calculet.tech2026-05-14T14:34:25+08:00- Removed references to CALRT from clip.h, mtmd-helper.cpp, mtmd-helper.h, mtmd.cpp, and mtmd.h. - Updated server.cpp to conditionally include CAL-LLM support. - Introduced option LLAMA_SERVER_USE_CAL_LLM in CMakeLists.txt to enable CAL-LLM integration. - Refactored server context to handle CAL-LLM requests and responses. - Updated image processing logic to remove CALRT-specific implementations. - Ensured compatibility with existing llama functionality while integrating new CAL-LLM features.
28b2003d6c90ef0fa26ab9e930ba3e22b355eb3d397800c772d44a7415640ec1669a691232939927cfwangbomengwangbomeng@calculet.tech2026-05-13T15:24:24+08:00wangbomengwangbomeng@calculet.tech2026-05-13T15:24:24+08:00try passkey and embedding
297800c772d44a7415640ec1669a691232939927cfdde8b8228081769c8d8f1e0751d47fa7875f8423wangbomengwangbomeng@calculet.tech2026-05-13T15:23:54+08:00wangbomengwangbomeng@calculet.tech2026-05-13T15:23:54+08:00try passkey and embedding
30dde8b8228081769c8d8f1e0751d47fa7875f8423fabcc7dac4b34d83fe430980bb364b81087c7ee9wangbomengwangbomeng@calculet.tech2026-05-12T10:46:08+08:00wangbomengwangbomeng@calculet.tech2026-05-12T10:46:08+08:00clip:add minicpm patch;
31fabcc7dac4b34d83fe430980bb364b81087c7ee91e9787962f20a7e82c71589ac1dd0585454e4bfewangbomengwangbomeng@calculet.tech2026-04-30T16:17:17+08:00wangbomengwangbomeng@calculet.tech2026-04-30T16:17:17+08:00calrt: fix encode dump
321e9787962f20a7e82c71589ac1dd0585454e4bfef584311c716dd078c4b737eebbfda97fe12b7274 1bec0db5a675ddc60fb793be2a746e8f3c8855dcwangbomengwangbomeng@calculet.tech2026-04-30T16:06:23+08:00wangbomengwangbomeng@calculet.tech2026-04-30T16:06:23+08:00Merge remote-tracking branch 'origin/dev-ben' into dev
33f584311c716dd078c4b737eebbfda97fe12b7274a339954cc2a145a80c5ea349e7896211916cbed9wangbomengwangbomeng@calculet.tech2026-04-30T16:05:15+08:00wangbomengwangbomeng@calculet.tech2026-04-30T16:05:15+08:00calrt: fix encode dump
341bec0db5a675ddc60fb793be2a746e8f3c8855dc465df20c3e1f3f58d0f5531c796e0941a49c32b0wangbomengwangbomeng@calculet.tech2026-04-30T16:02:59+08:00wangbomengwangbomeng@calculet.tech2026-04-30T16:02:59+08:00origin/dev-benbench: support 2 calbins
35465df20c3e1f3f58d0f5531c796e0941a49c32b0a339954cc2a145a80c5ea349e7896211916cbed9wangbomengwangbomeng@calculet.tech2026-04-30T09:08:12+08:00wangbomengwangbomeng@calculet.tech2026-04-30T09:08:12+08:00support one calbin
36a339954cc2a145a80c5ea349e7896211916cbed98e76fc330ad624d1768b412d7894e272dbe696f4 eeeef3f6d6a5aad9296323ca15d9d929745b8932wangbomengwangbomeng@calculet.tech2026-04-29T14:57:55+08:00wangbomengwangbomeng@calculet.tech2026-04-29T14:57:55+08:00tag: v0.3.0Merge branch 'dev' into dev-cli
378e76fc330ad624d1768b412d7894e272dbe696f4431ad6f1c787f1b47e9e7be14b2e1a7d8565485cwangbomengwangbomeng@calculet.tech2026-04-29T14:56:59+08:00wangbomengwangbomeng@calculet.tech2026-04-29T14:56:59+08:00finish dev of llama-cli
38eeeef3f6d6a5aad9296323ca15d9d929745b8932431ad6f1c787f1b47e9e7be14b2e1a7d8565485cwangbomengwangbomeng@calculet.tech2026-04-28T16:16:19+08:00wangbomengwangbomeng@calculet.tech2026-04-28T16:16:19+08:00add unknown time info: prefill and decode
39431ad6f1c787f1b47e9e7be14b2e1a7d8565485cefa1876bb8b2a449a55054a8c3745892f2c0f262wangbomengwangbomeng@calculet.tech2026-04-27T16:52:15+08:00wangbomengwangbomeng@calculet.tech2026-04-27T16:52:15+08:00update calrt_infer
40efa1876bb8b2a449a55054a8c3745892f2c0f2624a2a3181f1cea170105c771e6ba69d851b82b937wangbomengwangbomeng@calculet.tech2026-04-27T11:35:58+08:00wangbomengwangbomeng@calculet.tech2026-04-27T11:35:58+08:00tag: v0.2.2update version print
414a2a3181f1cea170105c771e6ba69d851b82b937e3ea5541259f35871c3e995331770aaca297e9ecwangbomengwangbomeng@calculet.tech2026-04-26T10:56:02+08:00wangbomengwangbomeng@calculet.tech2026-04-26T10:56:02+08:00add calrt-version
42e3ea5541259f35871c3e995331770aaca297e9ec0b8e43943846a1cd05c26c9186b9ef778ebae6e2wangbomengwangbomeng@calculet.tech2026-04-26T10:55:48+08:00wangbomengwangbomeng@calculet.tech2026-04-26T10:55:48+08:00add calrt-version
430b8e43943846a1cd05c26c9186b9ef778ebae6e25eb46fffbd2e260c540fb112314ad67bbb8f56b6wangbomengwangbomeng@calculet.tech2026-04-26T08:43:43+08:00wangbomengwangbomeng@calculet.tech2026-04-26T08:43:43+08:00to debug
445eb46fffbd2e260c540fb112314ad67bbb8f56b6005112c51e00d998b457337ca3dad53fc46e2d0bwangbomengwangbomeng@calculet.tech2026-04-24T16:24:51+08:00wangbomengwangbomeng@calculet.tech2026-04-24T16:24:51+08:00clip: add image_to_patches; calrt: vision copy data from bf16 to fp32
45005112c51e00d998b457337ca3dad53fc46e2d0b509670642e8090a1713f5d73590b549be827aaefwangbomengwangbomeng@calculet.tech2026-04-24T16:24:31+08:00wangbomengwangbomeng@calculet.tech2026-04-24T16:24:31+08:00clip: add image_to_patches; calrt: vision copy data from bf16 to fp32
46509670642e8090a1713f5d73590b549be827aaef5d219bfcef239c1e3ddd1baecc0514a9462ff6a8wangbomengwangbomeng@calculet.tech2026-04-24T09:40:34+08:00wangbomengwangbomeng@calculet.tech2026-04-24T09:40:34+08:00rm debug code
475d219bfcef239c1e3ddd1baecc0514a9462ff6a86787a53ab9b4897c5a7b51a4062d7b90e6697d35wangbomengwangbomeng@calculet.tech2026-04-24T09:31:32+08:00wangbomengwangbomeng@calculet.tech2026-04-24T09:31:32+08:00calrt: recover base = batch_index * dict_len
486787a53ab9b4897c5a7b51a4062d7b90e6697d35e1a76abf66178ea85656479a5e19aea7785c09f4wangbomengwangbomeng@calculet.tech2026-04-24T09:22:02+08:00wangbomengwangbomeng@calculet.tech2026-04-24T09:22:02+08:00calrt:input of multimodel from fp32 to bf16
49e1a76abf66178ea85656479a5e19aea7785c09f499dd308adb8488fa869bb669e3335e646a938effwangbomengwangbomeng@calculet.tech2026-04-22T16:29:22+08:00wangbomengwangbomeng@calculet.tech2026-04-22T16:29:22+08:00calrt:add CalcoreRT version print
5099dd308adb8488fa869bb669e3335e646a938eff06c7852e99e34c71e24e3ef6001cb54c72970e4dwangbomengwangbomeng@calculet.tech2026-04-21T15:50:01+08:00wangbomengwangbomeng@calculet.tech2026-04-21T15:50:01+08:00tag: v0.2.1server: fix max_seq_len
5106c7852e99e34c71e24e3ef6001cb54c72970e4d96b4e267e562f44e76ec6c965ee0fa361f0439bf ed62eb33447f993d578e156a3c7e38c91b420ecewangbomengwangbomeng@calculet.tech2026-04-21T15:06:01+08:00wangbomengwangbomeng@calculet.tech2026-04-21T15:06:01+08:00Merge branch 'dev-ot' into dev
52ed62eb33447f993d578e156a3c7e38c91b420ece015e988a271573a7c11e2a4b45084e1219d69d50wangbomengwangbomeng@calculet.tech2026-04-21T15:05:22+08:00wangbomengwangbomeng@calculet.tech2026-04-21T15:05:22+08:00try server-test:not success
53015e988a271573a7c11e2a4b45084e1219d69d5099962021f2dd1994dbbf90d5712f0ca2633f20f8wangbomengwangbomeng@calculet.tech2026-04-17T16:56:08+08:00wangbomengwangbomeng@calculet.tech2026-04-17T16:56:08+08:00little change
5496b4e267e562f44e76ec6c965ee0fa361f0439bf99962021f2dd1994dbbf90d5712f0ca2633f20f8wangbomengwangbomeng@calculet.tech2026-04-17T15:33:01+08:00wangbomengwangbomeng@calculet.tech2026-04-17T15:33:01+08:00merge dev-ot
5599962021f2dd1994dbbf90d5712f0ca2633f20f89af6682ef101c6d006c743c703cf150ac25f466fwangbomengwangbomeng@calculet.tech2026-04-17T14:48:33+08:00wangbomengwangbomeng@calculet.tech2026-04-17T14:48:33+08:00origin/dev-otcalrt: add onetoken slice and untile_prefill
569af6682ef101c6d006c743c703cf150ac25f466fe24f775ee26ea5249d5ff6cd23b8a0a17fe1438dwangbomengwangbomeng@calculet.tech2026-04-16T10:09:44+08:00wangbomengwangbomeng@calculet.tech2026-04-16T10:09:44+08:00args: add -ca
57e24f775ee26ea5249d5ff6cd23b8a0a17fe1438da206460e610aef995e36e50b33e8bd41fe477541wangbomengwangbomeng@calculet.tech2026-04-15T10:01:09+08:00wangbomengwangbomeng@calculet.tech2026-04-15T10:01:09+08:00calrt: add copy_data_to_ibuf for vision model
58795179beb5bba728e0ccf0a7e80a5ec5ef36e92a6d55eae7e76b768acfcec045b8e84cded8ae3b55Yunzhe Jiayunzhe@calculet.tech2026-04-14T16:03:22+08:00Yunzhe Jiayunzhe@calculet.tech2026-04-14T16:03:22+08:00origin/test-cicdadd ci yaml
59a206460e610aef995e36e50b33e8bd41fe477541d8b2aaa00049978cebac15041f960d75b552f97bwangbomengwangbomeng@calculet.tech2026-04-14T10:23:43+08:00wangbomengwangbomeng@calculet.tech2026-04-14T10:23:43+08:00calrt: support mixture of prefill_by_ids and prefill_by_embeds
60d8b2aaa00049978cebac15041f960d75b552f97bb6f29ec8140028619efaa50aa6e73146cecf631dwangbomengwangbomeng@calculet.tech2026-04-13T11:23:44+08:00wangbomengwangbomeng@calculet.tech2026-04-13T11:23:44+08:00server: fix context-shift
61b6f29ec8140028619efaa50aa6e73146cecf631d6d55eae7e76b768acfcec045b8e84cded8ae3b55wangbomengwangbomeng@calculet.tech2026-04-13T11:23:31+08:00wangbomengwangbomeng@calculet.tech2026-04-13T11:23:31+08:00server: fix context-shift
626d55eae7e76b768acfcec045b8e84cded8ae3b55af8055f5f033a730e7e1fe80185b18b7e9473966wangbomengwangbomeng@calculet.tech2026-04-10T14:58:56+08:00wangbomengwangbomeng@calculet.tech2026-04-10T14:58:56+08:00tag: v0.2.0add annotations
63af8055f5f033a730e7e1fe80185b18b7e94739666f54682df62b8ec6bb4154b99c47b5becda3291fwangbomengwangbomeng@calculet.tech2026-04-10T14:58:19+08:00wangbomengwangbomeng@calculet.tech2026-04-10T14:58:19+08:00add annotations
646f54682df62b8ec6bb4154b99c47b5becda3291f42a181674565411efa5c8b318bc77d6fd084df8a a2d5bf362d6c9a546f9deadf4f8975605a915d33Yunzhe Jiayunzhe@calculet.tech2026-04-10T09:07:20+08:00Yunzhe Jiayunzhe@calculet.tech2026-04-10T09:07:20+08:00Merge branch 'dev' of http://192.168.10.26:1000/yunzhe/llama.cpp into dev
6542a181674565411efa5c8b318bc77d6fd084df8abd3a57a11aa008e0b5dc2a30f82fcd3a02587208Yunzhe Jiayunzhe@calculet.tech2026-04-10T09:06:55+08:00Yunzhe Jiayunzhe@calculet.tech2026-04-10T09:06:55+08:00fix prompt_cache issue
66a2d5bf362d6c9a546f9deadf4f8975605a915d33bd3a57a11aa008e0b5dc2a30f82fcd3a02587208Bomeng Wangwangbomeng@calculet.tech2026-04-09T21:01:46+08:00Bomeng Wangwangbomeng@calculet.tech2026-04-09T21:01:46+08:00server: comment fixed time
67bd3a57a11aa008e0b5dc2a30f82fcd3a02587208fcfcdf80b01664a08fb12df879a05df3cc0eba70 74d437ab1456831573def7ca385a74d0ca460c37Yunzhe Jiayunzhe@calculet.tech2026-04-09T19:42:57+08:00Yunzhe Jiayunzhe@calculet.tech2026-04-09T19:42:57+08:00Merge branch 'dev' of http://192.168.10.26:1000/yunzhe/llama.cpp into dev
68fcfcdf80b01664a08fb12df879a05df3cc0eba700cd44929b442860dc0e5f88976108be82350f9dcYunzhe Jiayunzhe@calculet.tech2026-04-09T19:41:52+08:00Yunzhe Jiayunzhe@calculet.tech2026-04-09T19:41:52+08:00tmp fix one run >4K problem
6974d437ab1456831573def7ca385a74d0ca460c375813c943005d1ede551e67ee98e35aefd210ff22 0cd44929b442860dc0e5f88976108be82350f9dcwangbomengwangbomeng@calculet.tech2026-04-09T17:44:26+08:00wangbomengwangbomeng@calculet.tech2026-04-09T17:44:26+08:00Merge branch 'dev' of http://192.168.10.26:1000/yunzhe/llama.cpp into dev
705813c943005d1ede551e67ee98e35aefd210ff221f0d5bb0453f93ffa51faf68237c89251c2e0c40wangbomengwangbomeng@calculet.tech2026-04-09T17:38:57+08:00wangbomengwangbomeng@calculet.tech2026-04-09T17:38:57+08:00add --device-info
711f0d5bb0453f93ffa51faf68237c89251c2e0c40b8153b6b8f244b223c9f4c2940ae1f212634519ewangbomengwangbomeng@calculet.tech2026-04-09T17:38:37+08:00wangbomengwangbomeng@calculet.tech2026-04-09T17:38:37+08:00add --device-info
720cd44929b442860dc0e5f88976108be82350f9dcb8153b6b8f244b223c9f4c2940ae1f212634519eBomeng Wangwangbomeng@calculet.tech2026-04-09T15:08:50+08:00Bomeng Wangwangbomeng@calculet.tech2026-04-09T15:08:50+08:00rm calculet0.txt
73b8153b6b8f244b223c9f4c2940ae1f212634519e82842cfc3fa3d5c004c4540aeff8a81f38509beewangbomengwangbomeng@calculet.tech2026-04-09T15:07:02+08:00wangbomengwangbomeng@calculet.tech2026-04-09T15:07:02+08:00server: fix limit tokens to max_seq_len
7482842cfc3fa3d5c004c4540aeff8a81f38509bee622a22991089c6debd5f4b57738f3df3a0bda8b5wangbomengwangbomeng@calculet.tech2026-04-09T15:06:26+08:00wangbomengwangbomeng@calculet.tech2026-04-09T15:06:26+08:00server: fix limit tokens to max_seq_len
75622a22991089c6debd5f4b57738f3df3a0bda8b55f47a738ba56a37d0ff281a16212f41979568e35wangbomengwangbomeng@calculet.tech2026-04-09T10:36:31+08:00wangbomengwangbomeng@calculet.tech2026-04-09T10:36:31+08:00server: comment circle print
765f47a738ba56a37d0ff281a16212f41979568e35ecf01a17651cd7b1e8b85fd6075e83831463ee23 207ca38c751f7e6c251833eb11e89a2a0bb0fd07Yunzhe Jiayunzhe@calculet.tech2026-04-09T09:48:20+08:00Yunzhe Jiayunzhe@calculet.tech2026-04-09T09:48:20+08:00Merge branch 'dev' of http://192.168.10.26:1000/yunzhe/llama.cpp into dev
77ecf01a17651cd7b1e8b85fd6075e83831463ee2399d2078e89305035d87d966cee7987c7d50b3925Yunzhe Jiayunzhe@calculet.tech2026-04-09T09:47:55+08:00Yunzhe Jiayunzhe@calculet.tech2026-04-09T09:47:55+08:00fix n_parallel problem
78207ca38c751f7e6c251833eb11e89a2a0bb0fd0799d2078e89305035d87d966cee7987c7d50b3925wangbomengwangbomeng@calculet.tech2026-04-08T19:14:37+08:00wangbomengwangbomeng@calculet.tech2026-04-08T19:14:37+08:00delet extra printf
7999d2078e89305035d87d966cee7987c7d50b39259626239f5a9e568c9975d67dd6745fef90279a87Yunzhe Jiayunzhe@calculet.tech2026-04-08T19:01:16+08:00Yunzhe Jiayunzhe@calculet.tech2026-04-08T19:01:16+08:00fix buffer resize problem
809626239f5a9e568c9975d67dd6745fef90279a87da724f3e2a8af2c2537e3edcff41e9415ac66fefwangbomengwangbomeng@calculet.tech2026-04-08T17:47:04+08:00wangbomengwangbomeng@calculet.tech2026-04-08T17:47:04+08:00calrt: fix MapBuf
81da724f3e2a8af2c2537e3edcff41e9415ac66fef18814f9bf1a423ebe8f32d6043c9cbc5b21b0218Bomeng Wangwangbomeng@calculet.tech2026-04-08T16:58:32+08:00Bomeng Wangwangbomeng@calculet.tech2026-04-08T16:58:32+08:00calrt: add MapBuf
8218814f9bf1a423ebe8f32d6043c9cbc5b21b0218bbee06d6d851e3f84f75b578047967d7e0e4a8dcwangbomengwangbomeng@calculet.tech2026-04-08T13:55:52+08:00wangbomengwangbomeng@calculet.tech2026-04-08T13:55:52+08:00server : reset inference time
83bbee06d6d851e3f84f75b578047967d7e0e4a8dce126c21ed7b793b648d63533ea7ee30885f5a82awangbomengwangbomeng@calculet.tech2026-04-08T11:29:36+08:00wangbomengwangbomeng@calculet.tech2026-04-08T11:29:36+08:00calrt: add test time print
8439edc6ec2e9560ccb20a2b8e547f58102bb268c12006831fd6ea8a8e16518b81bcee29fb674676eaBomeng Wangwangbomeng@calculet.tech2026-04-01T14:06:50+08:00Bomeng Wangwangbomeng@calculet.tech2026-04-01T14:06:50+08:00tag: v0.1.1add t_incopy t_outcopy
852006831fd6ea8a8e16518b81bcee29fb674676ea41e6636ea1fb3884e41fb2023f92b0e7262ef09bwangbomengwangbomeng@calculet.tech2026-04-01T10:42:27+08:00wangbomengwangbomeng@calculet.tech2026-04-01T10:42:27+08:00tag: v0.1.0v0.1.0
86e126c21ed7b793b648d63533ea7ee30885f5a82a5603e050af07173589f3b8850121263d86da01fdwangbomengwangbomeng@calculet.tech2026-04-01T10:07:12+08:00wangbomengwangbomeng@calculet.tech2026-04-01T10:07:12+08:00sever: fix commit id
875603e050af07173589f3b8850121263d86da01fd1e8a16cd0e3ad3c9a0f8746852a0a38c6e52322ewangbomengwangbomeng@calculet.tech2026-04-01T05:57:55+08:00wangbomengwangbomeng@calculet.tech2026-04-01T05:57:55+08:00calrt: add t_incopy and t_outcopy
881e8a16cd0e3ad3c9a0f8746852a0a38c6e52322e893e52bda0fee99bbda1521ea733a760bc4201f1wangbomengwangbomeng@calculet.tech2026-04-01T10:14:22+08:00wangbomengwangbomeng@calculet.tech2026-04-01T10:14:22+08:00calrt: fix slice and add infer/copy time
89893e52bda0fee99bbda1521ea733a760bc4201f1398ecf71f5f574037ce33f84dd1576fa164909adBomeng Wangwangbomeng@calculet.tech2026-03-30T11:35:40+08:00Bomeng Wangwangbomeng@calculet.tech2026-03-30T11:35:40+08:00delete unnecessary commit id
90398ecf71f5f574037ce33f84dd1576fa164909ad6d5e83902bdd3142a8ff0d3db19f02a8619e95a0wangbomengwangbomeng@calculet.tech2026-03-27T02:01:23+08:00wangbomengwangbomeng@calculet.tech2026-03-27T02:01:23+08:00rm llama-server_old.cpp
916d5e83902bdd3142a8ff0d3db19f02a8619e95a0e4eaab700333490e80ebefc8d023f6c1adfcc312wangbomengwangbomeng@calculet.tech2026-03-27T00:57:11+08:00wangbomengwangbomeng@calculet.tech2026-03-27T00:57:11+08:00server: printf calrt commit ID
92e4eaab700333490e80ebefc8d023f6c1adfcc312aeacc23a883400db0ebffb50e23355ada054da66wangbomengwangbomeng@calculet.tech2026-03-25T09:03:55+08:00wangbomengwangbomeng@calculet.tech2026-03-25T09:03:55+08:00calrt: prefil to prefill
93aeacc23a883400db0ebffb50e23355ada054da66cbaa7492663630132adeddccd9fec5e44ffa3336wangbomengwangbomeng@calculet.tech2026-03-25T08:49:22+08:00wangbomengwangbomeng@calculet.tech2026-03-25T08:49:22+08:00calrt: add slice only on decode
94cbaa7492663630132adeddccd9fec5e44ffa33363c08ebf0e442cd730ddffd959ec676f9caf31c82wangbomengwangbomeng@calculet.tech2026-03-25T03:21:02+08:00wangbomengwangbomeng@calculet.tech2026-03-25T03:21:02+08:00reduce warning
953c08ebf0e442cd730ddffd959ec676f9caf31c82613bc2b9a9d7af92efbe42d876ffcdb244f7a00ewangbomengwangbomeng@calculet.tech2026-03-24T02:43:03+08:00wangbomengwangbomeng@calculet.tech2026-03-24T02:43:03+08:00calrt: comment LOG_ERROR in untile_decode_cpy
96613bc2b9a9d7af92efbe42d876ffcdb244f7a00e4e194ecc6c4b8ebe4c72d93f6055e3605dbfcefawangbomengwangbomeng@calculet.tech2026-03-24T02:30:12+08:00wangbomengwangbomeng@calculet.tech2026-03-24T02:30:12+08:00calrt: fix max_seq_len judgment
974e194ecc6c4b8ebe4c72d93f6055e3605dbfcefa16cbf81a731f5839a2f6b0eb9d8539826353748bwangbomengwangbomeng@calculet.tech2026-03-24T01:59:03+08:00wangbomengwangbomeng@calculet.tech2026-03-24T01:59:03+08:00calrt: fix ubatch.embd cpy
9816cbf81a731f5839a2f6b0eb9d8539826353748b8b4e17d990c23025148c4abadefe1010a0dd0734wangbomengwangbomeng@calculet.tech2026-03-23T09:38:21+08:00wangbomengwangbomeng@calculet.tech2026-03-23T09:38:21+08:00calrt: fix model_name of untile_cpy
998b4e17d990c23025148c4abadefe1010a0dd07346dacdd2cf9574214ca272a86674e7bd7ba21c4f0 41e6636ea1fb3884e41fb2023f92b0e7262ef09bwangbomengwangbomeng@calculet.tech2026-03-23T14:50:27+08:00wangbomengwangbomeng@calculet.tech2026-03-23T14:50:27+08:00merge origin dev
1006dacdd2cf9574214ca272a86674e7bd7ba21c4f03e6ad64467771224f072e5bad90e270951ba4cdfwangbomengwangbomeng@calculet.tech2026-03-23T14:42:28+08:00wangbomengwangbomeng@calculet.tech2026-03-23T14:42:28+08:00calrt: add multimodal
1013e6ad64467771224f072e5bad90e270951ba4cdfb37b7d9756feb3dac2db526eeab38b572a095d47wangbomengwangbomeng@calculet.tech2026-03-23T14:36:41+08:00wangbomengwangbomeng@calculet.tech2026-03-23T14:36:41+08:00calrt: add multimodal
10241e6636ea1fb3884e41fb2023f92b0e7262ef09bd0a1b2c758440720d42fa108b3444f54593daa73Bomeng Wangwangbomeng@calculet.tech2026-03-12T16:36:07+08:00Bomeng Wangwangbomeng@calculet.tech2026-03-12T16:36:07+08:00tag: v0.0.1server: correct the max_seq_len judgment
103d0a1b2c758440720d42fa108b3444f54593daa73b37b7d9756feb3dac2db526eeab38b572a095d47Bomeng Wangwangbomeng@calculet.tech2026-03-12T16:10:28+08:00Bomeng Wangwangbomeng@calculet.tech2026-03-12T16:10:28+08:00calrt: comment out dump
104b37b7d9756feb3dac2db526eeab38b572a095d4707817cefc14fa359dcecad0630defd3610f8f5c8 7ddafbdcb9426ae3af016925b8b596f11e8742d0Yunzhe Jiayunzhe@calculet.tech2026-03-10T16:32:31+08:00Yunzhe Jiayunzhe@calculet.tech2026-03-10T16:32:31+08:00Merge branch 'dev' of http://192.168.10.26:1000/yunzhe/llama.cpp into dev
10507817cefc14fa359dcecad0630defd3610f8f5c8b2c45d8ddefb7ec09435231c8ec5b736fabf9c91Yunzhe Jiayunzhe@calculet.tech2026-03-10T16:32:07+08:00Yunzhe Jiayunzhe@calculet.tech2026-03-10T16:32:07+08:00fix kv issue
1067ddafbdcb9426ae3af016925b8b596f11e8742d0059009eeaf20a569abfbac5eb37dad3bb9376428wangbomengwangbomeng@calculet.tech2026-03-10T07:13:04+08:00wangbomengwangbomeng@calculet.tech2026-03-10T07:13:04+08:00calrt: commented out cal_ctx->encode(batch)
107059009eeaf20a569abfbac5eb37dad3bb93764281b073661d1311b5300173993b9f31ee7241d8606wangbomengwangbomeng@calculet.tech2026-03-10T07:04:22+08:00wangbomengwangbomeng@calculet.tech2026-03-10T07:04:22+08:00calrt: add multimodel
1081b073661d1311b5300173993b9f31ee7241d86066295273c5866b9249bacdedbfc3a31877e6e6a85 b2c45d8ddefb7ec09435231c8ec5b736fabf9c91wangbomengwangbomeng@calculet.tech2026-03-10T06:58:35+08:00wangbomengwangbomeng@calculet.tech2026-03-10T06:58:35+08:00Merge branch 'dev' of http://192.168.20.229:1000/yunzhe/llama.cpp into dev
1096295273c5866b9249bacdedbfc3a31877e6e6a85c6ea3f0d2468e3b8eab5de80bacee7fe4c9dd3d7wangbomengwangbomeng@calculet.tech2026-03-10T06:54:41+08:00wangbomengwangbomeng@calculet.tech2026-03-10T06:54:41+08:00arg: update hf_repo url to https://download.calculet.tech:9443
110b2c45d8ddefb7ec09435231c8ec5b736fabf9c91c6ea3f0d2468e3b8eab5de80bacee7fe4c9dd3d7Bomeng Wangwangbomeng@calculet.tech2026-03-09T15:03:26+08:00Bomeng Wangwangbomeng@calculet.tech2026-03-09T15:03:26+08:00clart: past_kv_len = n_tokens
111c6ea3f0d2468e3b8eab5de80bacee7fe4c9dd3d7f8fba66b657c51a1df1efc7267bfea9f6140cebfwangbomengwangbomeng@calculet.tech2026-03-09T03:26:47+08:00wangbomengwangbomeng@calculet.tech2026-03-09T03:26:47+08:00CMakeLists: STATIC to SHARED
112f8fba66b657c51a1df1efc7267bfea9f6140cebfdb4a61d2234c2f457b42ecc3f7219a1e1a35508aYunzhe Jiayunzhe@calculet.tech2026-03-06T15:06:36+08:00Yunzhe Jiayunzhe@calculet.tech2026-03-06T07:05:39+08:00fix calrt_install include path
113db4a61d2234c2f457b42ecc3f7219a1e1a35508ae672e02ca1a582f9e3a9d011d9e1b4fea4942a25Yunzhe Jiayunzhe@calculet.tech2026-03-06T06:59:58+08:00Yunzhe Jiayunzhe@calculet.tech2026-03-06T06:59:58+08:00adapt to host-rt install structure
114e672e02ca1a582f9e3a9d011d9e1b4fea4942a2527b82f64252779ad160a3fa76bd378c39421de36 7a3a9a0e76bb5f530beafcfe52e87e388e93117aYunzhe Jiayunzhe@calculet.tech2026-03-06T11:47:09+08:00Yunzhe Jiayunzhe@calculet.tech2026-03-06T11:47:09+08:00Merge branch 'dev' of http://192.168.10.26:1000/yunzhe/llama.cpp into dev
11527b82f64252779ad160a3fa76bd378c39421de360bd95719fed2f0247c8d0199dd897d24157dfae6Yunzhe Jiayunzhe@calculet.tech2026-03-06T11:42:45+08:00Yunzhe Jiayunzhe@calculet.tech2026-03-06T11:45:03+08:00adapt to calrt_objs target
1167a3a9a0e76bb5f530beafcfe52e87e388e93117a583c387efaf7bc030574e554088de79d09a27fbdwangbomengwangbomeng@calculet.tech2026-03-05T07:24:59+08:00wangbomengwangbomeng@calculet.tech2026-03-05T07:24:59+08:00calrt_infer: fix past_kv_len
117583c387efaf7bc030574e554088de79d09a27fbd0edc14a60a2d5596f71d728c655e3523da924d89wangbomengwangbomeng@calculet.tech2026-02-27T02:24:56+08:00wangbomengwangbomeng@calculet.tech2026-02-27T02:24:56+08:00caserver: lim generated tokens to n_ctx_per_seq
1180edc14a60a2d5596f71d728c655e3523da924d89aa0a17d719326a63e0d6fea60aa0ced44e7abc33wangbomengwangbomeng@calculet.tech2026-02-27T01:57:39+08:00wangbomengwangbomeng@calculet.tech2026-02-27T01:57:39+08:00caserver: limit generated tokens to n_ctx_per_seq
119aa0a17d719326a63e0d6fea60aa0ced44e7abc33fa332e773da681f3e77d6ffe83a87edb557f758bwangbomengwangbomeng@calculet.tech2026-02-27T01:40:11+08:00wangbomengwangbomeng@calculet.tech2026-02-27T01:40:11+08:00calrt: add bf16_to_fp32
120fa332e773da681f3e77d6ffe83a87edb557f758baa54a7c637f2c92d924a5f35386975b1c27de898wangbomengwangbomeng@calculet.tech2026-02-27T01:31:42+08:00wangbomengwangbomeng@calculet.tech2026-02-27T01:31:42+08:00Add bf16_to_fp32
121aa54a7c637f2c92d924a5f35386975b1c27de89884f2c0b5f334a7ab77347c6779b6027a951d0ff0wangbomengwangbomeng@calculet.tech2026-02-27T01:30:36+08:00wangbomengwangbomeng@calculet.tech2026-02-27T01:30:36+08:00Add bf16_to_fp32
12284f2c0b5f334a7ab77347c6779b6027a951d0ff00bd95719fed2f0247c8d0199dd897d24157dfae6wangbomengwangbomeng@calculet.tech2026-02-09T01:46:36+08:00wangbomengwangbomeng@calculet.tech2026-02-09T01:46:36+08:00add bf16_to_fp32 in untile_decode/prefill_cpy
1230bd95719fed2f0247c8d0199dd897d24157dfae6365eb0b719e30b0f9e348d854f11c9c3f954a96eYunzhe Jiayunzhe@calculet.tech2026-02-07T10:25:06+08:00Yunzhe Jiayunzhe@calculet.tech2026-02-07T10:25:06+08:00fix get_model_by_name
124365eb0b719e30b0f9e348d854f11c9c3f954a96e886cd1fb3b217efcf51c46209fcb694022346797Yunzhe Jiayunzhe@calculet.tech2026-02-07T09:05:17+08:00Yunzhe Jiayunzhe@calculet.tech2026-02-07T09:05:17+08:00comment out pos buffer file
125886cd1fb3b217efcf51c46209fcb6940223467979328f21378275f2354a19c3af907332b216ab492Yunzhe Jiayunzhe@calculet.tech2026-02-06T17:36:54+08:00Yunzhe Jiayunzhe@calculet.tech2026-02-06T17:36:54+08:00fix compile problem
1269328f21378275f2354a19c3af907332b216ab49249fea70a20f550bfdb59d0ffe041b54919c22447Yunzhe Jiayunzhe@calculet.tech2026-01-28T15:32:36+08:00Yunzhe Jiayunzhe@calculet.tech2026-01-28T15:32:36+08:00add prefill-only mode
12749fea70a20f550bfdb59d0ffe041b54919c2244774a118df31b6a4acc8876b2a2db90d2395cc6ec3Yunzhe Jiayunzhe@calculet.tech2026-01-28T09:11:16+08:00Yunzhe Jiayunzhe@calculet.tech2026-01-28T09:11:16+08:00adapt to new calbin test_case
12874a118df31b6a4acc8876b2a2db90d2395cc6ec398e8f9acfe87cd93646482cb2d942b8a132f0852Yunzhe Jiayunzhe@calculet.tech2026-01-13T14:32:47+08:00Yunzhe Jiayunzhe@calculet.tech2026-01-13T14:32:47+08:00sync with calrt update
12998e8f9acfe87cd93646482cb2d942b8a132f08528c8152a04036bb0d4756a4e08e54f514dae4e64dYunzhe Jiayunzhe@calculet.tech2025-12-10T00:20:57+08:00Yunzhe Jiayunzhe@calculet.tech2025-12-10T00:20:57+08:00add timestamp
1308c8152a04036bb0d4756a4e08e54f514dae4e64d5d2387931528fbe04aa8d66de935147efb65ca4dYunzhe Jiayunzhe@calculet.tech2025-12-09T14:40:04+08:00Yunzhe Jiayunzhe@calculet.tech2025-12-09T14:40:28+08:00adapt to calrt csr
1315d2387931528fbe04aa8d66de935147efb65ca4de6285a748657add8d974b78ca920c5b41753ee12Yunzhe Jiayunzhe@calculet.tech2025-12-08T16:40:34+08:00Yunzhe Jiayunzhe@calculet.tech2025-12-08T16:40:54+08:00add timestamp for one decode process
132e6285a748657add8d974b78ca920c5b41753ee126b1ecdbbe3cc72ed8888be38f9e934cfd79b9fbbYunzhe Jiayunzhe@calculet.tech2025-11-27T16:58:35+08:00Yunzhe Jiayunzhe@calculet.tech2025-11-27T16:58:35+08:00adapt to calrt
1336b1ecdbbe3cc72ed8888be38f9e934cfd79b9fbb240d9bb75241c91896afe125f2fc74698a7395f3Yunzhe Jiayunzhe@calculet.tech2025-11-25T17:06:14+08:00Yunzhe Jiayunzhe@calculet.tech2025-11-25T17:06:14+08:00add kv_tensor for pld test
134240d9bb75241c91896afe125f2fc74698a7395f37eeb241aad429a68fb8cfa3ba85d000fb2348cd2Yunzhe Jiayunzhe@calculet.tech2025-11-24T12:03:00+08:00Yunzhe Jiayunzhe@calculet.tech2025-11-24T12:03:00+08:00adapt to calrt modification
1357eeb241aad429a68fb8cfa3ba85d000fb2348cd2dfb6250031a25058dbc3757e8a7ec75d90dcb48aYunzhe Jiayunzhe@calculet.tech2025-11-21T09:20:01+08:00Yunzhe Jiayunzhe@calculet.tech2025-11-21T09:20:01+08:00fix byte_size error
136dfb6250031a25058dbc3757e8a7ec75d90dcb48ae31bce25e85dc7fdf0a0e69fd0e58ba4a2ffaf3eYunzhe Jiayunzhe@calculet.tech2025-11-18T10:28:00+08:00Yunzhe Jiayunzhe@calculet.tech2025-11-18T10:43:01+08:00add elf in CMakelist
137e31bce25e85dc7fdf0a0e69fd0e58ba4a2ffaf3e6444e227c36df16199c40d8cead2244fe014f377Yunzhe Jiayunzhe@calculet.tech2025-11-07T11:52:17+08:00Yunzhe Jiayunzhe@calculet.tech2025-11-07T11:52:17+08:00dump src tensor
1386444e227c36df16199c40d8cead2244fe014f377d81d7b190ff5f21325167936496dfe5ab48b922eYunzhe Jiayunzhe@calculet.tech2025-11-07T09:19:11+08:00Yunzhe Jiayunzhe@calculet.tech2025-11-07T09:19:11+08:00test golden case
139d81d7b190ff5f21325167936496dfe5ab48b922ee492f5dc8cec17529f53ef2bbdf7b502107f8148 1f5accb8d0056e6099cd5b772b1cb787dd590a13Yunzhe Jiayunzhe@calculet.tech2025-11-04T14:43:32+08:00Yunzhe Jiayunzhe@calculet.tech2025-11-04T15:04:07+08:00Merge branch 'master' into dev
140e492f5dc8cec17529f53ef2bbdf7b502107f81482e05afd22b3d77c7013b11eee88e88b17188f21aYunzhe Jiayunzhe@calculet.tech2025-11-04T14:09:59+08:00Yunzhe Jiayunzhe@calculet.tech2025-11-04T14:09:59+08:00redo untile logic
1411f5accb8d0056e6099cd5b772b1cb787dd590a132759ccdb4adc8568add4316780d5e675519b0775Noah99681487+NoahOksuz@users.noreply.github.com2025-11-04T05:04:59ZGitHubnoreply@github.com2025-11-03T21:04:59-08:00Fix garbled output with REPACK at high thread counts (#16956)
1422759ccdb4adc8568add4316780d5e675519b0775c5023daf607c578d6344c628eb7da18ac3d92d32Aman Guptaamangupta052@gmail.com2025-11-04T10:53:48+08:00GitHubnoreply@github.com2025-11-04T10:53:48+08:00CUDA: avoid mul + bias fusion when doing fusion (#16935)
143c5023daf607c578d6344c628eb7da18ac3d92d32e7da30b584dc1f2ee0414c4a1298ce64eef97e8dlhezlih@qti.qualcomm.com2025-11-03T11:47:57-08:00GitHubnoreply@github.com2025-11-03T11:47:57-08:00opencl: support imrope (#16914)
144e7da30b584dc1f2ee0414c4a1298ce64eef97e8ded8aa63320393512bdcfe4b05b5ae01ba91888e1Aleksander Grygieraleksander.grygier@gmail.com2025-11-03T18:53:26+01:00GitHubnoreply@github.com2025-11-03T18:53:26+01:00fix: Viewing multiple PDF attachments (#16974)
145ed8aa63320393512bdcfe4b05b5ae01ba91888e148bd26501b08a3f0bff1249db47f313641f7bebbDaniel Beveniusdaniel.bevenius@gmail.com2025-11-03T18:01:59+01:00GitHubnoreply@github.com2025-11-03T18:01:59+01:00model-conversion : pass config to from_pretrained (#16963)
14648bd26501b08a3f0bff1249db47f313641f7bebb622cd010ff4a65bef67edbf2f9bf4707c01f98f7Georgi Gerganovggerganov@gmail.com2025-11-03T15:38:23+02:00GitHubnoreply@github.com2025-11-03T14:38:23+01:00server : add props.model_alias (#16943)
147622cd010ff4a65bef67edbf2f9bf4707c01f98f7070ff4d5356083d60b807bb34d36b31c3653a29etheo77186theo77186@users.noreply.github.com2025-11-03T14:29:11+01:00GitHubnoreply@github.com2025-11-03T14:29:11+01:00ggml: CUDA: add head size 72 for flash-attn (#16962)
148070ff4d5356083d60b807bb34d36b31c3653a29ebf7b0c9725ed0c406c5debaea022ec7258430634Xuan-Son Nguyenson@huggingface.co2025-11-03T11:11:18+01:00GitHubnoreply@github.com2025-11-03T11:11:18+01:00mtmd: add --image-min/max-tokens (#16921)
149bf7b0c9725ed0c406c5debaea022ec7258430634fcfce040e816c542f374ad51461b3561d73c4bc9Xuan-Son Nguyenson@huggingface.co2025-11-03T10:25:55+01:00GitHubnoreply@github.com2025-11-03T10:25:55+01:00mtmd: pad mask for qwen2.5vl (#16954)
150fcfce040e816c542f374ad51461b3561d73c4bc9ee3a5a10adf9e83722d1914dddc56a0623ececafJinyang Hehejinyang@loongson.cn2025-11-03T14:40:02+08:00GitHubnoreply@github.com2025-11-03T08:40:02+02:00ggml : LoongArch fixes (#16958)
151ee3a5a10adf9e83722d1914dddc56a0623ececaf7e994168b1ccc12337ba8de939c4fd466107c1fbOlivier Chafikolivier.chafik@gmail.com2025-11-03T05:33:56ZGitHubnoreply@github.com2025-11-03T07:33:56+02:00sync: minja (glm 4.6 & minmax m2 templates) (#16949)
1527e994168b1ccc12337ba8de939c4fd466107c1fbbcfa87622ae46be6345a8e3dfdbdc5ba5414042bshani-fs0556787439@gmail.com2025-11-03T03:35:33+02:00GitHubnoreply@github.com2025-11-03T09:35:33+08:00SYCL: optimized repeat_back kernel (3× fewer asm instructions, 2× faster)Feature/sycl repeat back opt (#16869)
153bcfa87622ae46be6345a8e3dfdbdc5ba5414042ba2054e3a8ff0da3978a4acc18c349ff58554d336Sascha Rogmann59577610+srogmann@users.noreply.github.com2025-11-03T00:41:08+01:00GitHubnoreply@github.com2025-11-03T00:41:08+01:00feat(webui): improve LaTeX rendering with currency detection (#16508)
154a2054e3a8ff0da3978a4acc18c349ff58554d336dd5286805004db1f9ac3176a1cbbfe373bdda0f8Shagun Bera141054835+notV3NOM@users.noreply.github.com2025-11-03T04:40:30+05:30GitHubnoreply@github.com2025-11-03T00:10:30+01:00test-backend-ops : fix segfault in moe-expert-reduce test in support mode and coverage (#16936)
155dd5286805004db1f9ac3176a1cbbfe373bdda0f86b9a52422bac0f50dd8f1f8386744fa3ce9783bfSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-11-02T23:11:21+01:00GitHubnoreply@github.com2025-11-02T23:11:21+01:00ci : disable failing riscv cross build (#16952)
1566b9a52422bac0f50dd8f1f8386744fa3ce9783bf2f966b8ed87514e74bb96592217226cb6a6974ddZhiyong Wang85110830+ravenouse@users.noreply.github.com2025-11-02T13:08:04-08:00GitHubnoreply@github.com2025-11-02T22:08:04+01:00model: add Janus Pro for image understanding (#16906)
1572f966b8ed87514e74bb96592217226cb6a6974ddcd5e3b57541ecc52421130742f4d89acbcf77cd4Georgi Gerganovggerganov@gmail.com2025-11-02T22:21:48+02:00GitHubnoreply@github.com2025-11-02T21:21:48+01:00clip : use FA (#16837)
158cd5e3b57541ecc52421130742f4d89acbcf77cd487c9efc3b297b8a498716b1db3d061842e6fc85bGeorgi Gerganovggerganov@gmail.com2025-11-02T18:14:04+02:00GitHubnoreply@github.com2025-11-02T18:14:04+02:00server : support unified cache across slots (#16736)
15987c9efc3b297b8a498716b1db3d061842e6fc85b76af40aaaad78c42faecd8016a88362c788b84b0Aldehir Rojashello@alde.dev2025-11-02T08:56:28-06:00GitHubnoreply@github.com2025-11-02T16:56:28+02:00common : move gpt-oss reasoning processing to init params (#16937)
16076af40aaaad78c42faecd8016a88362c788b84b07db35a7958a943be1693879f42d166f152613979Adrian Lundberg47256989+alundb@users.noreply.github.com2025-11-02T10:28:37+01:00GitHubnoreply@github.com2025-11-02T11:28:37+02:00docs: remove llama_sampler_accept reference in sampling sample usage (#16920)
1617db35a7958a943be1693879f42d166f152613979a864132ba546b6385922226480ee4e392aaa065cmnehete3233429707+mnehete32@users.noreply.github.com2025-11-02T08:42:57+05:30GitHubnoreply@github.com2025-11-02T11:12:57+08:00CUDA: add FLOOR, CEIL, ROUND, TRUNC unary ops (#16917)
162a864132ba546b6385922226480ee4e392aaa065cd38d9f0877a5872daa3c5f06fb9a86376bf15d50Aaron Teoaaron.teo1@ibm.com2025-11-02T08:48:46+08:00GitHubnoreply@github.com2025-11-02T08:48:46+08:00devops: fix failing s390x docker build (#16918)
163d38d9f0877a5872daa3c5f06fb9a86376bf15d507fd205a8e8832ead273f62c5fd81d2b8aa910535Aaron Teoaaron.teo1@ibm.com2025-11-02T08:48:23+08:00GitHubnoreply@github.com2025-11-02T08:48:23+08:00ggml: add s390x cpu-feats (#16774)
1647fd205a8e8832ead273f62c5fd81d2b8aa9105352f68ce7cfd20e9e7098514bf730e5389b7bba908Georgi Gerganovggerganov@gmail.com2025-11-02T00:15:31+02:00GitHubnoreply@github.com2025-11-02T00:15:31+02:00scripts : add script to bench models (#16894)
1652f68ce7cfd20e9e7098514bf730e5389b7bba908e4a71599e5846110159955dec0008eb4aa24222bPascaladmin@serveurperso.com2025-11-01T19:49:51+01:00GitHubnoreply@github.com2025-11-01T19:49:51+01:00webui: auto-refresh /props on inference start to resync model metadata (#16784)
166e4a71599e5846110159955dec0008eb4aa24222bdd5e8cab512c7752392b6e51a6f118f348fb3f16Pascaladmin@serveurperso.com2025-11-01T17:14:54+01:00GitHubnoreply@github.com2025-11-01T17:14:54+01:00webui: add HTML/JS preview support to MarkdownContent with sandboxed iframe (#16757)
167dd5e8cab512c7752392b6e51a6f118f348fb3f16cf659bbb8ef9eb048e5153a27cf787fd83c05560Adrien Gallouëtangt@huggingface.co2025-11-01T16:52:17+01:00GitHubnoreply@github.com2025-11-01T16:52:17+01:00vendor : update cpp-httplib to 0.27.0 (#16846)
168cf659bbb8ef9eb048e5153a27cf787fd83c05560d8b860a219c2415faac8cc0e50b48b4aa11e3b64Xuan-Son Nguyenson@huggingface.co2025-11-01T15:51:36+01:00GitHubnoreply@github.com2025-11-01T15:51:36+01:00mtmd: refactor preprocessing + support max/min pixels (#16878)
169d8b860a219c2415faac8cc0e50b48b4aa11e3b641ae74882f8f6755e44dff8f23f3abdc5b53ab7c1Aleksander Grygieraleksander.grygier@gmail.com2025-11-01T15:35:57+01:00GitHubnoreply@github.com2025-11-01T15:35:57+01:00Add a setting to display message generation statistics (#16901)
1701ae74882f8f6755e44dff8f23f3abdc5b53ab7c1961660b8c395afb8903f61ace69396a158ea0cb4Jaromír Hradílekjhradilek@gmail.com2025-11-01T15:02:57+01:00GitHubnoreply@github.com2025-11-01T15:02:57+01:00webui: recognize AsciiDoc files as valid text files (#16850)
171961660b8c395afb8903f61ace69396a158ea0cb474fef4129fa58f6277c243777a4894371479dbadSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-11-01T11:01:42+01:00GitHubnoreply@github.com2025-11-01T11:01:42+01:00common : allow --system-prompt-file for diffusion-cli (#16903)
17274fef4129fa58f6277c243777a4894371479dbad5d8bb900bc7daa84bfa7bb1d25ab7e32394919f3Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-11-01T08:55:25+01:00GitHubnoreply@github.com2025-11-01T09:55:25+02:00codeowners : update after refactor (#16905)
1735d8bb900bc7daa84bfa7bb1d25ab7e32394919f32e76e013600cb0d51ccf158571ca1d0502952a07Jeff Bolzjbolz@nvidia.com2025-11-01T00:52:14-05:00GitHubnoreply@github.com2025-11-01T06:52:14+01:00vulkan: Fix multi_add invalid descriptor usage (#16899)
1742e76e013600cb0d51ccf158571ca1d0502952a07d3dc9dd898be805c23a408cc36daed5b3bf29221Jeff Bolzjbolz@nvidia.com2025-11-01T00:45:28-05:00GitHubnoreply@github.com2025-11-01T06:45:28+01:00vulkan: fuse mul_mat+add and mul_mat_id+add_id (#16868)
175d3dc9dd898be805c23a408cc36daed5b3bf29221bea04522ff1a0d8559ccfd353aa018dcfbb608ccOliver Simonsosimons@nvidia.com2025-11-01T06:13:26+01:00GitHubnoreply@github.com2025-11-01T13:13:26+08:00CUDA: Remove unneded bias/gate dims in fused mmvq (#16858)
176bea04522ff1a0d8559ccfd353aa018dcfbb608cc0de0a01576772032008a689afc4d7c80685074c4Piotr Wilkin (ilintar)piotr.wilkin@syndatis.com2025-10-31T23:40:23+01:00GitHubnoreply@github.com2025-10-31T23:40:23+01:00refactor : llama-model.cpp (#16252)
1770de0a01576772032008a689afc4d7c80685074c4e58d585604bbbbbc24f8149effe444621b5191c6Piotr Wilkin (ilintar)piotr.wilkin@syndatis.com2025-10-31T21:20:47+01:00GitHubnoreply@github.com2025-10-31T21:20:47+01:00model : Minimax M2 (#16831)
178e58d585604bbbbbc24f8149effe444621b5191c631c511a968348281e11d590446bb815048a1e912Giuseppe Scrivanogscrivan@redhat.com2025-10-31T21:20:07+01:00GitHubnoreply@github.com2025-10-31T21:20:07+01:00model : add Granite Hybrid nano types (#16896)
17931c511a968348281e11d590446bb815048a1e9126d39015a744b47bb7643ea73f412e6d64677f305Johannes Gäßlerjohannesg@5d6.de2025-10-31T15:57:19+01:00GitHubnoreply@github.com2025-10-31T15:57:19+01:00CUDA: Volta tensor core support for MMF (#16843)
1806d39015a744b47bb7643ea73f412e6d64677f3054146d6a1a6228711a487a1e3e9ddd120f8d027d7Georgi Gerganovggerganov@gmail.com2025-10-31T16:25:50+02:00Georgi Gerganovggerganov@gmail.com2025-10-31T16:26:28+02:00sync : ggml
1814146d6a1a6228711a487a1e3e9ddd120f8d027d78da3c0e200a586f768ada6f38745acb01380174cAman Guptaamangupta052@gmail.com2025-10-31T20:05:07+08:00GitHubnoreply@github.com2025-10-31T20:05:07+08:00CUDA: add expert reduce kernel (#16857)
1828da3c0e200a586f768ada6f38745acb01380174cc22473b580807929fd9e3a3344a48e8cfbe6c88fGeorgi Gerganovggerganov@gmail.com2025-10-31T13:50:33+02:00GitHubnoreply@github.com2025-10-31T13:50:33+02:00batch : fix consistency checks for the input positions (#16890)
183c22473b580807929fd9e3a3344a48e8cfbe6c88f0f715b4e759acceccb9f437cfd2a988fff85514aGeorgi Gerganovggerganov@gmail.com2025-10-31T10:54:19+02:00GitHubnoreply@github.com2025-10-31T10:54:19+02:00server : don't print user inputs to console (#16871)
1840f715b4e759acceccb9f437cfd2a988fff85514ad2d931f173b8a736b08999436e9259aafddec718Daniel Beveniusdaniel.bevenius@gmail.com2025-10-31T09:51:26+01:00GitHubnoreply@github.com2025-10-31T09:51:26+01:00server : fix typos in server.cpp comments [no ci] (#16883)
185d2d931f173b8a736b08999436e9259aafddec7182976b0374d36609b0429dd6ce48807e2ad39a7c2Jeff Bolzjbolz@nvidia.com2025-10-31T02:34:47-05:00GitHubnoreply@github.com2025-10-31T08:34:47+01:00vulkan: disable spirv-opt for rope shaders (#16872)
1862976b0374d36609b0429dd6ce48807e2ad39a7c2d2a2673dd1c86a01ab010e18b13b8bb959968c48Masato Nakasakamasato.nakasaka@intel.com2025-10-31T16:18:59+09:00GitHubnoreply@github.com2025-10-31T08:18:59+01:00vulkan: Fix crash when FP16 mul_mat accumulation is not supported (#16796)
187d2a2673dd1c86a01ab010e18b13b8bb959968c4813002a08960e51a76c4d696165b5d7638d2f2b99Ruben Ortlampicard12@live.de2025-10-31T08:14:49+01:00GitHubnoreply@github.com2025-10-31T08:14:49+01:00vulkan: fix shmem overrun in mmq id shader (#16873)
1882e05afd22b3d77c7013b11eee88e88b17188f21ac20c0a5c0c44179f5267a262546624854b1203d1Yunzhe Jiayunzhe@calculet.tech2025-10-31T14:06:06+08:00Yunzhe Jiayunzhe@calculet.tech2025-10-31T14:06:06+08:00add pcie support
18913002a08960e51a76c4d696165b5d7638d2f2b996eb208d17ea29bb60295d9a2b5e7122dfb8f4b55l3utterflygc.pthzfoldr@gmail.com2025-10-31T12:46:31+08:00GitHubnoreply@github.com2025-10-30T21:46:31-07:00ggml-hexagon: respect input size when getting/setting tensor data (#16836)
1906eb208d17ea29bb60295d9a2b5e7122dfb8f4b559984cbb61d12491d604484fb38b678fa15064061Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-31T00:34:27+01:00GitHubnoreply@github.com2025-10-31T00:34:27+01:00ci : enable free-disk-space on cuda docker build (#16877)
1919984cbb61d12491d604484fb38b678fa15064061ce18efeaf1234bd08f25bc08f88ef95cf5d9e51flhezlih@qti.qualcomm.com2025-10-30T16:00:20-07:00GitHubnoreply@github.com2025-10-30T16:00:20-07:00opencl: fix boundary handling for mul_mm (#16875)
192ce18efeaf1234bd08f25bc08f88ef95cf5d9e51f16724b5b6836a2d4b8936a5824d2ff27c52b4517RodriMorabullerwins@gmail.com2025-10-30T23:15:03+01:00GitHubnoreply@github.com2025-10-30T23:15:03+01:00convert : update transformers requirements (#16866)
19316724b5b6836a2d4b8936a5824d2ff27c52b4517b52edd25586fabb70f0c21b274473b307cf14499chansikparkchansik.park@gmail.com2025-10-30T14:22:23-04:00GitHubnoreply@github.com2025-10-30T20:22:23+02:00server : bump request URI max length to 32768 (#16862)
194b52edd25586fabb70f0c21b274473b307cf14499517b7170e1a4d733583c4b07c5b7a49acc05911cGeorgi Gerganovggerganov@gmail.com2025-10-30T18:42:57+02:00GitHubnoreply@github.com2025-10-30T18:42:57+02:00server : remove n_past (#16818)
195517b7170e1a4d733583c4b07c5b7a49acc05911c835e918d8428f5119927d7150bf5a26176dedda0Max Krasnyanskymaxk@qti.qualcomm.com2025-10-30T09:06:13-07:00GitHubnoreply@github.com2025-10-30T09:06:13-07:00cpu: introduce chunking for repack matmuls and enable matmul-id chunking on ARM64 (#16833)
196835e918d8428f5119927d7150bf5a26176dedda0d261223d24e97f2df50220e4a5b7f0adb69bba81Shagun Bera141054835+notV3NOM@users.noreply.github.com2025-10-30T21:17:31+05:30GitHubnoreply@github.com2025-10-30T17:47:31+02:00common: fix typo in cli help text (#16864)
197d261223d24e97f2df50220e4a5b7f0adb69bba81dcca0d3ab840ebe9b2ccd4719033d408eeb758d7JJJYmmm92386084+JJJYmmm@users.noreply.github.com2025-10-30T23:19:14+08:00GitHubnoreply@github.com2025-10-30T16:19:14+01:00model: add support for qwen3vl series (#16780)
198dcca0d3ab840ebe9b2ccd4719033d408eeb758d7bacddc049a00786df44e682262f6e298742bfbc3Max Krasnyanskymaxk@qti.qualcomm.com2025-10-30T05:26:05-07:00GitHubnoreply@github.com2025-10-30T14:26:05+02:00cpu: introduce chunking for flash attention (#16829)
199bacddc049a00786df44e682262f6e298742bfbc3229bf686287d18f82c44e89888cc662145ecfdb4Tianyue-Zhaozhaotianyue@outlook.com2025-10-30T07:18:50-04:00GitHubnoreply@github.com2025-10-30T12:18:50+01:00model: Add support for CogVLM model (#15002)
200229bf686287d18f82c44e89888cc662145ecfdb4d7395115baf395b75a73a17b0b796e746e468da9Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-30T08:56:28+01:00GitHubnoreply@github.com2025-10-30T08:56:28+01:00cuda : fix argsort with 64k+ rows (#16849)
201d7395115baf395b75a73a17b0b796e746e468da9052df28b0e3ca0398e5928c61c7de40254317894Jan Boonjan.boon@kaetemi.be2025-10-30T14:30:58+08:00GitHubnoreply@github.com2025-10-30T08:30:58+02:00llama : use std::abs instead of abs (#16853)
202052df28b0e3ca0398e5928c61c7de402543178948b11deea4663f29d3e042ce1056ba643264cd5f1Jeff Bolzjbolz@nvidia.com2025-10-30T01:27:41-05:00GitHubnoreply@github.com2025-10-30T07:27:41+01:00vulkan: Handle argsort with a large number of rows (#16851)
2038b11deea4663f29d3e042ce1056ba643264cd5f1b9ce94017729465895402cbcfffb51fa926c15e3Oliver Simonsosimons@nvidia.com2025-10-30T04:34:15+01:00GitHubnoreply@github.com2025-10-30T11:34:15+08:00Hide latency of bias and gate-loading (#16847)
204b9ce94017729465895402cbcfffb51fa926c15e33464bdac37027c5e9661621fc75ffcef3c19c6efJeff Bolzjbolz@nvidia.com2025-10-29T15:13:10-05:00GitHubnoreply@github.com2025-10-29T15:13:10-05:00vulkan: Fuse rope+set_rows (#16769)
2053464bdac37027c5e9661621fc75ffcef3c19c6efe3af5563bd049141e036b50f843196db33d23e97Xuan-Son Nguyenson@huggingface.co2025-10-29T20:11:39+01:00GitHubnoreply@github.com2025-10-29T20:11:39+01:00llama: fix ASAN error with M-RoPE (#16848)
206e3af5563bd049141e036b50f843196db33d23e9710fcc41290e233788f5a4215314156e8e023eb92Xuan-Son Nguyenson@huggingface.co2025-10-29T18:09:18+01:00GitHubnoreply@github.com2025-10-29T18:09:18+01:00llama: store mrope data in KV cell (#16825)
20710fcc41290e233788f5a4215314156e8e023eb92bcf5bda6f5df559565d11d7c8e8295c1159a85ecJeff Bolzjbolz@nvidia.com2025-10-29T08:44:29-05:00GitHubnoreply@github.com2025-10-29T14:44:29+01:00vulkan: Update topk_moe fusion to handle gpt's late softmax (#16656)
208bcf5bda6f5df559565d11d7c8e8295c1159a85ec3eb2be1ca5f37480aeb16102970d9e65f43347feRuben Ortlampicard12@live.de2025-10-29T14:39:03+01:00GitHubnoreply@github.com2025-10-29T14:39:03+01:00Vulkan MMQ Integer Dot Refactor and K-Quant support (#16536)
2093eb2be1ca5f37480aeb16102970d9e65f43347fee41bcce8f0b53032a1fed275cd253e931c041cf6Max Krasnyanskymaxk@qti.qualcomm.com2025-10-29T06:29:12-07:00GitHubnoreply@github.com2025-10-29T06:29:12-07:00Hexagon Op queue & dispatch optimizations (#16820)
210e41bcce8f0b53032a1fed275cd253e931c041cf6144a4ce824b6bd0e48d62009d10cae1daf5308dbAman Guptaamangupta052@gmail.com2025-10-29T21:11:53+08:00GitHubnoreply@github.com2025-10-29T21:11:53+08:00CUDA: use fastdiv in set-rows (#16834)
211144a4ce824b6bd0e48d62009d10cae1daf5308dbf549b0007dbdd683215820f7229ce180a12b191dSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-29T14:09:50+01:00GitHubnoreply@github.com2025-10-29T14:09:50+01:00vendor : sync minja (#16500)
212f549b0007dbdd683215820f7229ce180a12b191d9a3ea685b937c0f0cbfda2e50004ea54bf187512Jeff Bolzjbolz@nvidia.com2025-10-29T03:53:04-05:00GitHubnoreply@github.com2025-10-29T09:53:04+01:00vulkan: Call ggml_vk_buffer_write_2d from ggml_vk_buffer_copy (#16793)
2139a3ea685b937c0f0cbfda2e50004ea54bf187512338074c383c81366320d176d83b94b0a567ee0c2Aman Guptaamangupta052@gmail.com2025-10-29T15:55:06+08:00GitHubnoreply@github.com2025-10-29T15:55:06+08:00CUDA: Fix bug in topk-moe for gpt-oss (#16821)
214338074c383c81366320d176d83b94b0a567ee0c2851553ea6b24cb39fd5fd188b437d777cb411de8YaelLogicy0548591250@gmail.com2025-10-29T08:14:39+02:00GitHubnoreply@github.com2025-10-29T14:14:39+08:00sycl: add RMS_NORM_BACK operation support (#16808)
215851553ea6b24cb39fd5fd188b437d777cb411de885a7d8677bf2200981e52f744a21d5267964ffcfYaelGitAccount38328157276@mby.co.il2025-10-28T21:10:28+02:00GitHubnoreply@github.com2025-10-28T20:10:28+01:00cuda: add SET operation support (#16804)
21685a7d8677bf2200981e52f744a21d5267964ffcfa8ca18b4b815a2abdbecb958ee5f4c542d69aac7Georgi Gerganovggerganov@gmail.com2025-10-28T20:19:44+02:00GitHubnoreply@github.com2025-10-28T20:19:44+02:00memory : remove KV cache size padding (#16812)
217a8ca18b4b815a2abdbecb958ee5f4c542d69aac78284efc35c909217ee1b9a06903245d808ac2283Georgi Gerganovggerganov@gmail.com2025-10-28T19:41:43+02:00GitHubnoreply@github.com2025-10-28T19:41:43+02:00llama-bench : clarify benchmarked parts of the computation (#16823)
2188284efc35c909217ee1b9a06903245d808ac22831c1409e13169d0ced4b1b4d39fb5b268b7525091l3utterflygc.pthzfoldr@gmail.com2025-10-28T23:16:20+08:00GitHubnoreply@github.com2025-10-28T08:16:20-07:00initialise buffer.device in ggml_hexagon_session (#16816)
2191c1409e13169d0ced4b1b4d39fb5b268b75250917a0e900e3615fa46c074a7fdf900b47d3c0a1c7eSam Malayek12037535+SamMalayek@users.noreply.github.com2025-10-28T03:51:41-07:00GitHubnoreply@github.com2025-10-28T12:51:41+02:00embedding: add raw option for --embd-output-format (#16541)
2207a0e900e3615fa46c074a7fdf900b47d3c0a1c7e280d97be9660e7a5feaa28a6e7a299bc73dd83fcJohannes Gäßlerjohannesg@5d6.de2025-10-28T11:23:54+01:00GitHubnoreply@github.com2025-10-28T11:23:54+01:00llama: consistent ctx <-> buf order for KV cache (#16746)
221280d97be9660e7a5feaa28a6e7a299bc73dd83fc3479efd112b3910d2d008ad08fe4cb4895605903Aldehir Rojashello@alde.dev2025-10-28T03:37:52-05:00GitHubnoreply@github.com2025-10-28T09:37:52+01:00grammar : support array references in json schema (#16792)
2223479efd112b3910d2d008ad08fe4cb4895605903463bbf20bfbe0da10ac58984d4d62bed5a49362aChenguang Li757486878@qq.com2025-10-28T10:54:53+08:00GitHubnoreply@github.com2025-10-28T10:54:53+08:00CANN: Improve device ID handling and aclnnArange checks (#16752)
223463bbf20bfbe0da10ac58984d4d62bed5a49362aad8d36beffd791db10c94eb9e964afb891e3ca55Aman Guptaamangupta052@gmail.com2025-10-28T10:31:21+08:00GitHubnoreply@github.com2025-10-28T10:31:21+08:00CUDA: add unused vars to mmvf and mmvq (#16807)
224ad8d36beffd791db10c94eb9e964afb891e3ca55c053e18a66dd95dc340aa61317877c2a41d4e3cftamarPaltamarp3385@gmail.com2025-10-28T03:50:33+02:00GitHubnoreply@github.com2025-10-28T09:50:33+08:00sycl: add SSM_CONV operation support (#16800)
225c053e18a66dd95dc340aa61317877c2a41d4e3cfe1ab0848037c9f9bfe68b3e1cee9ee375e1018a3Yuri Khrustalevykhrustalev@users.noreply.github.com2025-10-27T18:54:01-04:00GitHubnoreply@github.com2025-10-27T23:54:01+01:00chat: Add LFM2 tool handling (#16763)
226e1ab0848037c9f9bfe68b3e1cee9ee375e1018a35a4ff43e7dd049e35942bc3d12361dab2f155544Xuan-Son Nguyenson@huggingface.co2025-10-27T23:12:16+01:00GitHubnoreply@github.com2025-10-27T23:12:16+01:00mtmd : fix idefics3 preprocessing (#16806)
2275a4ff43e7dd049e35942bc3d12361dab2f15554410640e31aab0819f31c1e1f2d008b019ee737232Diego Devesaslarengh@gmail.com2025-10-27T13:51:28-07:00GitHubnoreply@github.com2025-10-27T21:51:28+01:00llama : disable pipeline parallelism if compute buffer allocation fails (#16748)
22810640e31aab0819f31c1e1f2d008b019ee73723280d28f104c0c3e61c11d6af073642b246a9fc19cAclyaclysia@gmail.com2025-10-27T21:50:22+01:00GitHubnoreply@github.com2025-10-27T21:50:22+01:00ggml : fix interpolate with align-corners and ne=1 (#16700)
22980d28f104c0c3e61c11d6af073642b246a9fc19cc55d53acec864f64afa1ba92972203dce1bf88f5Johannes Gäßlerjohannesg@5d6.de2025-10-27T21:39:49+01:00GitHubnoreply@github.com2025-10-27T21:39:49+01:00HIP: fix AMDGPU_TARGETS, update documentation (#16803)
230c55d53acec864f64afa1ba92972203dce1bf88f5945501f5ea4b8ca56b181ecb035e9ee3fb31f432Xuan-Son Nguyenson@huggingface.co2025-10-27T16:02:58+01:00GitHubnoreply@github.com2025-10-27T16:02:58+01:00model : add LightOnOCR-1B model (#16764)
231945501f5ea4b8ca56b181ecb035e9ee3fb31f43275cbdd3fce38ea12d50cd19e73a069aa5dbbd5faJohannes Gäßlerjohannesg@5d6.de2025-10-27T09:17:31+01:00GitHubnoreply@github.com2025-10-27T09:17:31+01:00llama: fix leaked buffers for mmap + split files (#16765)
23275cbdd3fce38ea12d50cd19e73a069aa5dbbd5fa2b9bd9bf4e759c05db629ec1c391dc8aeaa71887Aman Guptaamangupta052@gmail.com2025-10-27T09:25:10+08:00GitHubnoreply@github.com2025-10-27T09:25:10+08:00test-backend-ops: print failed tests at the end (#16785)
2332b9bd9bf4e759c05db629ec1c391dc8aeaa7188759fc1ec8e83b14354c1a3a8acf8c5c2cbf9af42ftamarPaltamarp3385@gmail.com2025-10-27T03:20:24+02:00GitHubnoreply@github.com2025-10-27T09:20:24+08:00sycl: add ROLL operation support (#16665)
23459fc1ec8e83b14354c1a3a8acf8c5c2cbf9af42f75d33b9302f84a5b89f82205d2bcd8def5a64e0ashani-fs0556787439@gmail.com2025-10-27T03:19:50+02:00GitHubnoreply@github.com2025-10-27T09:19:50+08:00sycl: add REPEAT_BACK operation support (#16734)
23575d33b9302f84a5b89f82205d2bcd8def5a64e0a3470a5c891dcc94363e492a3760af92b6b07241cAman Guptaamangupta052@gmail.com2025-10-27T09:06:16+08:00GitHubnoreply@github.com2025-10-27T09:06:16+08:00CUDA: support for weight clamp in top-k norm (#16702)
2363470a5c891dcc94363e492a3760af92b6b07241cbd562fe4f7bd55625511d5f9d639c4fb1db1d440Aclyaclysia@gmail.com2025-10-26T23:19:03+01:00GitHubnoreply@github.com2025-10-26T23:19:03+01:00ggml-alloc : make gallocr prefer chunks that allow memory reuse (#16788)
237bd562fe4f7bd55625511d5f9d639c4fb1db1d440bbac6a26b2bd7f7c1f0831cb1e7b52734c66673bSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-26T21:31:41+01:00GitHubnoreply@github.com2025-10-26T21:31:41+01:00cuda : use fast copy when src and dst are of different type and contiguous (#16789)
238bbac6a26b2bd7f7c1f0831cb1e7b52734c66673b73a48c9790d320476b3e5ef75bda09f2f8269e6eleejetleejet714@gmail.com2025-10-27T02:13:31+08:00GitHubnoreply@github.com2025-10-26T19:13:31+01:00ggml: fix cuda kernel launch configuration for k_compute_batched_ptrs to support large batch (#16744)
23973a48c9790d320476b3e5ef75bda09f2f8269e6ef696428ce8e4d16c17acbffeaa7feac3b0fb9061Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-26T17:21:23+01:00GitHubnoreply@github.com2025-10-26T17:21:23+01:00convert : enable expert group selection for all models with it (#16691)
240f696428ce8e4d16c17acbffeaa7feac3b0fb90617cce4f8158f0c4c88d8dadd4c23d33938127b897Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-26T17:20:32+01:00GitHubnoreply@github.com2025-10-26T17:20:32+01:00graph : add clamping to ffn_moe_weights_sum to avoid div-by-zero (#16655)
2417cce4f8158f0c4c88d8dadd4c23d33938127b8978d8862829cd770fc9c0c7a726ed162de824ff5eaSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-26T16:08:52+01:00GitHubnoreply@github.com2025-10-26T16:08:52+01:00model : set res->t_embd in SmallThinker models (#16782)
2428d8862829cd770fc9c0c7a726ed162de824ff5eaf77c13b91f4d25754b6a0b857f98a6bc922a0aa7amirai2189905406+amirai21@users.noreply.github.com2025-10-26T14:01:20+02:00GitHubnoreply@github.com2025-10-26T13:01:20+01:00docs : add Jamba to Text-only models list (#16778)
243f77c13b91f4d25754b6a0b857f98a6bc922a0aa73cfa9c3f125763305b4226bc032f1954f08990dcAman Guptaamangupta052@gmail.com2025-10-26T19:28:04+08:00GitHubnoreply@github.com2025-10-26T19:28:04+08:00CUDA: General GEMV fusion (#16715)
2443cfa9c3f125763305b4226bc032f1954f08990dc5d195f17bc60eacc15cfb929f9403cf29ccdf419Gilad S.7817232+giladgd@users.noreply.github.com2025-10-26T06:37:38+02:00GitHubnoreply@github.com2025-10-26T05:37:38+01:00vulkan: deduplicate Microsoft Direct3D12 devices (#16689)
2455d195f17bc60eacc15cfb929f9403cf29ccdf419226f295f4dd92ad714533adc5497afed5fa88bb8Galunidkarolek1231456@gmail.com2025-10-25T20:41:36+02:00GitHubnoreply@github.com2025-10-25T20:41:36+02:00convert : handle mmproj filename/path properly (#16760)
246226f295f4dd92ad714533adc5497afed5fa88bb8f90b4a8efe4466215c51155a5c22c1ae6207da23Shunta Saitoshunta.saito@gmail.com2025-10-25T19:26:27+09:00GitHubnoreply@github.com2025-10-25T12:26:27+02:00model : set res->t_embd in PLaMo2 models (#16766)
247f90b4a8efe4466215c51155a5c22c1ae6207da238423d019318b446640bb620e4ce80066d8530f05Giuseppe Scrivanogscrivan@redhat.com2025-10-25T10:59:54+02:00GitHubnoreply@github.com2025-10-25T10:59:54+02:00vulkan: delete dead code (#16732)
2488423d019318b446640bb620e4ce80066d8530f055cca2542ac3f3f86831d32bce744d08fc2b353b0Jeff Bolzjbolz@nvidia.com2025-10-25T00:04:12-05:00GitHubnoreply@github.com2025-10-25T07:04:12+02:00vulkan: Optimize SSM_SCAN (#16645)
2495cca2542ac3f3f86831d32bce744d08fc2b353b055945d2ef51b93821d4b6f4a9b994393344a90dbcompiladegit@compilade.net2025-10-24T20:52:00-04:00GitHubnoreply@github.com2025-10-24T20:52:00-04:00convert : avoid dequantizing mxfp4 for GPT-OSS (#16756)
25055945d2ef51b93821d4b6f4a9b994393344a90db0bcb40b48c6fc6f17ba9672625e526ab2574344bleejetleejet714@gmail.com2025-10-25T03:39:37+08:00GitHubnoreply@github.com2025-10-24T21:39:37+02:00ggml: fix CUDA grid launch condition for large block_nums.y in binbcast (#16742)
2510bcb40b48c6fc6f17ba9672625e526ab2574344b69e9ff010309a1155d704cf9320bdb3aaf4160caAman Guptaamangupta052@gmail.com2025-10-24T20:46:19+08:00GitHubnoreply@github.com2025-10-24T20:46:19+08:00CUDA: use CUB for arbitary size argsort (#16754)
25269e9ff010309a1155d704cf9320bdb3aaf4160ca5a91109a5d7dab5d7adc40bedb397ede99a705b1Florian Badieflorianbadie@odrling.xyz2025-10-24T14:10:29+02:00GitHubnoreply@github.com2025-10-24T14:10:29+02:00webui: support q URL parameter (#16728)
2535a91109a5d7dab5d7adc40bedb397ede99a705b1f8f071faddf32ea09f4234edb6e809b380a9ee26Daniel Beveniusdaniel.bevenius@gmail.com2025-10-24T12:02:02+02:00GitHubnoreply@github.com2025-10-24T12:02:02+02:00model-conversion : add trust_remote_code for orig model run [no ci] (#16751)
254f8f071faddf32ea09f4234edb6e809b380a9ee260bf47a1dbba4d36f2aff4e8c34b06210ba34e688compiladegit@compilade.net2025-10-23T16:31:41-04:00GitHubnoreply@github.com2025-10-23T16:31:41-04:00convert : handle pre-quantized models (#14810)
2550bf47a1dbba4d36f2aff4e8c34b06210ba34e688dd62dcfab97e420949519fd0eac9fca7bf97e635Johannes Gäßlerjohannesg@5d6.de2025-10-23T21:30:17+02:00GitHubnoreply@github.com2025-10-23T21:30:17+02:00server: add memory breakdown print (#16740)
256dd62dcfab97e420949519fd0eac9fca7bf97e635d0660f237a5c31771a3d6d1030ebe3e0c409ba92Julien Denize40604584+juliendenize@users.noreply.github.com2025-10-23T15:54:46+02:00GitHubnoreply@github.com2025-10-23T15:54:46+02:00convert : Make mistral-common dependency optional (#16738)
257d0660f237a5c31771a3d6d1030ebe3e0c409ba92fe6a9882acf5c02f96624ed8f80144100d7006cbXuan-Son Nguyenson@huggingface.co2025-10-23T15:00:49+02:00GitHubnoreply@github.com2025-10-23T15:00:49+02:00mtmd-cli : allow using --jinja (#16718)
258fe6a9882acf5c02f96624ed8f80144100d7006cb061f0eff02dc9a82f7bd850db3bd70b8a0b5e87aPrajwal B Mehendarkarprajwal.b.mehendarkar@ibm.com2025-10-23T17:07:31+05:30GitHubnoreply@github.com2025-10-23T19:37:31+08:00Manually link -lbsd to resolve flock symbol on AIX (#16610)
259061f0eff02dc9a82f7bd850db3bd70b8a0b5e87a8cf6b42d467d05fa7d9776d2bcc69974ecce6900Aman Guptaamangupta052@gmail.com2025-10-23T19:14:06+08:00GitHubnoreply@github.com2025-10-23T19:14:06+08:00ggml-cuda: use passed ops instead of hardcoded ops (#16712)
2608cf6b42d467d05fa7d9776d2bcc69974ecce69009de9672adb0f4ca4e39483ac3ffed52b3f70a55dmatteomatteo.serva@gmail.com2025-10-23T11:32:24+02:00GitHubnoreply@github.com2025-10-23T12:32:24+03:00server : send partial stop string when <EOG> is reached (#15007)
2619de9672adb0f4ca4e39483ac3ffed52b3f70a55d63d2fc46e17a06be5b4b5823a5ada088317f1f0aMatthew Michelmatthew.michel@intel.com2025-10-22T20:05:15-05:00GitHubnoreply@github.com2025-10-23T09:05:15+08:00sycl: use async memory allocation to fix crashes during graph recording (#16644)
26263d2fc46e17a06be5b4b5823a5ada088317f1f0aa2e0088d9242bd9e57f8b852b05a6e47843b5a45Max Krasnyanskymaxk@qti.qualcomm.com2025-10-22T13:47:09-07:00GitHubnoreply@github.com2025-10-22T13:47:09-07:00Add experimental ggml-hexagon backend for the Hexagon NPU (#16547)
2639b9201f65a22c02cee8e300f58f480a58859122719a5a3edfd306516cc419679d69d6435943b6816Pascaladmin@serveurperso.com2025-10-22T16:58:23+02:00GitHubnoreply@github.com2025-10-22T16:58:23+02:00webui: introduce OpenAI-compatible model selector in JSON payload (#16562)
26419a5a3edfd306516cc419679d69d6435943b6816d8eaa26e4d9228df3aa46a930db60c8eaab67c1bsirus20x6sirus20x6@users.noreply.github.com2025-10-22T05:14:14-05:00GitHubnoreply@github.com2025-10-22T12:14:14+02:00ggml : Leverage the existing GGML_F32_VEC helpers to vectorize ggml_vec_set_f32 for faster fills (#16522)
265d8eaa26e4d9228df3aa46a930db60c8eaab67c1b9285325ce0174631c3cd6121d56084adc4ef2d8fAclyaclysia@gmail.com2025-10-22T12:01:22+02:00GitHubnoreply@github.com2025-10-22T12:01:22+02:00tests : fix test-thread-safety when compiling with multiple backends (#16699)
2669285325ce0174631c3cd6121d56084adc4ef2d8f03792ad93609fc67e41041c6347d9aa14e5e0d74Aman Guptaamangupta052@gmail.com2025-10-22T12:33:08+08:00GitHubnoreply@github.com2025-10-22T12:33:08+08:00CUDA: fix bug in topk-moe softmax (#16711)
26703792ad93609fc67e41041c6347d9aa14e5e0d7451d1a8c997bd2629ef211a30208058ea87a30982Aman Guptaamangupta052@gmail.com2025-10-21T22:40:38+08:00GitHubnoreply@github.com2025-10-21T22:40:38+08:00CUDA: topk-moe: add optional parameter for gpt-oss (#16649)
26851d1a8c997bd2629ef211a30208058ea87a309824926419c4d74a1cf724e7163d937eb72f36e7b26Johannes Gäßlerjohannesg@5d6.de2025-10-21T15:27:53+02:00GitHubnoreply@github.com2025-10-21T15:27:53+02:00CUDA: better error for FA kernel with 0 occupancy (#16643)
2694926419c4d74a1cf724e7163d937eb72f36e7b266ea37f57391d27736c35cd3c20c1f990b7952b74Aman Guptaamangupta052@gmail.com2025-10-21T16:43:14+08:00GitHubnoreply@github.com2025-10-21T16:43:14+08:00ggml: add ggml_can_fuse_subgraph (#16662)
2706ea37f57391d27736c35cd3c20c1f990b7952b74fb349848f387f355450c3187556e71e6d32c145flhezlih@qti.qualcomm.com2025-10-20T22:26:17-07:00GitHubnoreply@github.com2025-10-20T22:26:17-07:00opencl: fix warnings and clean up profiling (#16688)
271fb349848f387f355450c3187556e71e6d32c145f6de8ed75196c7cd98c1f34bbf3a7452451ba8ac2Jeff Bolzjbolz@nvidia.com2025-10-20T22:16:08-05:00GitHubnoreply@github.com2025-10-20T22:16:08-05:00vulkan: Handle FA with all -inf mask values (#16447)
2726de8ed75196c7cd98c1f34bbf3a7452451ba8ac284bf3c677857279037adf67cdcfd89eaa4ca9281YehuditEy8703470@gmail.com2025-10-21T01:21:12+03:00GitHubnoreply@github.com2025-10-21T00:21:12+02:00sycl : add PAD_REFLECT_D1 operator support (#16145)
27384bf3c677857279037adf67cdcfd89eaa4ca9281c9c1972e2c2cc6a771fcc145bfa138700179f961Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-20T21:38:20+02:00GitHubnoreply@github.com2025-10-20T21:38:20+02:00model : add BailingMoeV2 support (#16063)
274c9c1972e2c2cc6a771fcc145bfa138700179f961b617cfd2896edd592a36ebbc041817eb030a1005Aleksander Grygieraleksander.grygier@gmail.com2025-10-20T19:49:02+02:00GitHubnoreply@github.com2025-10-20T19:49:02+02:00Handle legacy 'context' attachments (#16687)
275b617cfd2896edd592a36ebbc041817eb030a100579068501fac9a74cca7129a8e5a8281b410a853eDiego Devesaslarengh@gmail.com2025-10-20T05:53:50-07:00GitHubnoreply@github.com2025-10-20T14:53:50+02:00ggml-alloc : fix leak when reusing a tensor with a larger size (#16679)
27679068501fac9a74cca7129a8e5a8281b410a853e0e4a0cf2fae667d3efcf52f2f52398779d986b1dAleksander Grygieraleksander.grygier@gmail.com2025-10-20T14:21:12+02:00GitHubnoreply@github.com2025-10-20T14:21:12+02:00Prevent premature submission on IME input (#16673)
2770e4a0cf2fae667d3efcf52f2f52398779d986b1d13f2cfad4170c096c51a02c24a6a158cb47f1480Aleksander Grygieraleksander.grygier@gmail.com2025-10-20T13:29:14+02:00GitHubnoreply@github.com2025-10-20T13:29:14+02:00Import/Export UX improvements (#16619)
27813f2cfad4170c096c51a02c24a6a158cb47f148006332e28672356b964d6dfc2ba4657e20581cd43Aleksander Grygieraleksander.grygier@gmail.com2025-10-20T12:41:13+02:00GitHubnoreply@github.com2025-10-20T12:41:13+02:00Enable per-conversation loading states to allow having parallel conversations (#16327)
27906332e28672356b964d6dfc2ba4657e20581cd4372d53e6c4decee8b339e49aed8cc0e234b9639dctakuya kodamaotegami@clear-code.com2025-10-20T16:27:09+08:00GitHubnoreply@github.com2025-10-20T11:27:09+03:00llama-batch: fix build fails with `-Werror=missing-braces` (#16614)
28072d53e6c4decee8b339e49aed8cc0e234b9639dc2330de7b847ca84eac766df372c604c26db72747Ron Evansron@hybridgroup.com2025-10-20T10:20:04+02:00GitHubnoreply@github.com2025-10-20T11:20:04+03:00readme: update bindings (#16651)
2812330de7b847ca84eac766df372c604c26db727477062dd8460685d6700ed7621e50a22c6f3400ca3safranowithbsh155762@gmail.com2025-10-20T11:08:32+03:00GitHubnoreply@github.com2025-10-20T11:08:32+03:00SYCL: Add support for FLOOR,CEIL,ROUND and TRUNC unary operators (#16613)
2827062dd8460685d6700ed7621e50a22c6f3400ca30398752dd450dfabdd1b9e289f6364c2600f6ab5takuya kodamaotegami@clear-code.com2025-10-20T15:44:21+08:00GitHubnoreply@github.com2025-10-20T10:44:21+03:00llama-context: only warn on pooling_type when user specified (#16674)
283c20c0a5c0c44179f5267a262546624854b1203d162425b3a7e8b169a3bbe6ce8c7324513a7fcee8cYunzhe Jiayunzhe@calculet.tech2025-10-20T14:41:08+08:00Yunzhe Jiayunzhe@calculet.tech2025-10-20T14:41:08+08:00kv: turn on common-prefix mtmd: fix output size bug
2840398752dd450dfabdd1b9e289f6364c2600f6ab54f73d0a95120687e2c527739f771330a5271259aGiuseppe Scrivanogscrivan@redhat.com2025-10-19T23:54:31+02:00GitHubnoreply@github.com2025-10-19T23:54:31+02:00model : add Granite Hybrid types (#16635)
2854f73d0a95120687e2c527739f771330a5271259acec5edbcaec69bbf6d5851cabce4ac148be41701Aaron Teoaaron.teo1@ibm.com2025-10-20T05:06:39+08:00GitHubnoreply@github.com2025-10-19T23:06:39+02:00ci : fix binaries release failure for s390x (binaries may not work yet) (#16664)
286cec5edbcaec69bbf6d5851cabce4ac148be41701fcb235b46618921cbd826acd49b553b5302233aaSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-19T14:03:25+02:00GitHubnoreply@github.com2025-10-19T14:03:25+02:00ci : avoid manual updates of docs/ops.md (#16663)
287fcb235b46618921cbd826acd49b553b5302233aa55754bebd5d570960cde9c0ba991a0b2991f6b1eAaron Teoaaron.teo1@ibm.com2025-10-19T18:37:47+08:00GitHubnoreply@github.com2025-10-19T18:37:47+08:00ci: include s390x release binaries (#16648)
28855754bebd5d570960cde9c0ba991a0b2991f6b1eee09828cb057460b369576410601a3a09279e23cAman Guptaamangupta052@gmail.com2025-10-19T15:37:12+08:00GitHubnoreply@github.com2025-10-19T10:37:12+03:00CODEOWNERS: update for ggml-cuda/mmf (#16660)
289ee09828cb057460b369576410601a3a09279e23ce56abd2098dd2e2b0804691b93c13b48ae421627Johannes Gäßlerjohannesg@5d6.de2025-10-18T14:47:32+02:00GitHubnoreply@github.com2025-10-18T14:47:32+02:00HIP: fix GPU_TARGETS (#16642)
290e56abd2098dd2e2b0804691b93c13b48ae42162738355c6c8e43204e11a22daa7483082c0ff01e71Jeff Bolzjbolz@nvidia.com2025-10-18T05:22:57-05:00GitHubnoreply@github.com2025-10-18T12:22:57+02:00vulkan: Implement topk_moe fused shader, ported from CUDA (#16641)
29138355c6c8e43204e11a22daa7483082c0ff01e7181387858f1fbcc1acedbd308486e1016618ca8f8Aman Guptaamangupta052@gmail.com2025-10-18T17:52:53+08:00GitHubnoreply@github.com2025-10-18T11:52:53+02:00CUDA: use registers instead of smem in topk-moe (#16647)
29281387858f1fbcc1acedbd308486e1016618ca8f866b0dbcb2d462e7b70ba5a69ee8c3899ac2efb1cShawn Gushawngu@qti.qualcomm.com2025-10-17T17:55:32-07:00GitHubnoreply@github.com2025-10-17T17:55:32-07:00opencl: transposed gemm/gemv moe kernel with mxfp4,f32 (#16602)
29366b0dbcb2d462e7b70ba5a69ee8c3899ac2efb1c41386cf365d894134ee0813d15e2f5d76f6a4d8eJohannes Gäßlerjohannesg@5d6.de2025-10-17T17:41:09+02:00GitHubnoreply@github.com2025-10-17T17:41:09+02:00llama-model: fix insonsistent ctxs <-> bufs order (#16581)
29441386cf365d894134ee0813d15e2f5d76f6a4d8e3d4e86bbeb15f487d6da6174ba6191b7c212cc25Radoslav Gerganovrgerganov@gmail.com2025-10-17T18:02:52+03:00GitHubnoreply@github.com2025-10-17T18:02:52+03:00rpc : report actual free memory (#16616)
2953d4e86bbeb15f487d6da6174ba6191b7c212cc25342c728d031d50673feded797520a44127d73379Giuseppe Scrivanogscrivan@redhat.com2025-10-17T14:23:47+02:00GitHubnoreply@github.com2025-10-17T14:23:47+02:00vulkan: Add State Space Model (SSM) Operations Support (#16463)
296342c728d031d50673feded797520a44127d73379ababae7e1ec3e9cfdab0322ee55ea3389e82a4d5muggle-stackpromuggle@qq.com2025-10-17T18:01:23+08:00GitHubnoreply@github.com2025-10-17T13:01:23+03:00ggml : fix SpaceMit IME array out-of-bounds in task assignment (#16629)
297ababae7e1ec3e9cfdab0322ee55ea3389e82a4d5b19491599d4d42c606601d75e95c1d1de3291f8ePascaladmin@serveurperso.com2025-10-17T10:35:03+02:00GitHubnoreply@github.com2025-10-17T10:35:03+02:00webui: reorganize settings layout (#16607)
298b19491599d4d42c606601d75e95c1d1de3291f8e9ad4f1931ee0f3b41d9355245ef744786aaae0aaJeff Bolzjbolz@nvidia.com2025-10-17T02:31:04-05:00GitHubnoreply@github.com2025-10-17T09:31:04+02:00vulkan: fix debug build (add_rms_len/data not found) (#16624)
2999ad4f1931ee0f3b41d9355245ef744786aaae0aa79967ec596c0dacfd2251b085a57e79df292b1ccIlia Ilmeriliailmer@users.noreply.github.com2025-10-17T02:33:58-04:00GitHubnoreply@github.com2025-10-17T09:33:58+03:00metal : add `CONV_TRANSPOSE_2D` (#16542)
30079967ec596c0dacfd2251b085a57e79df292b1ccceff6bb253dd306f5404d7ccb3f11fadafe71b52Olivier Chafikolivier.chafik@gmail.com2025-10-17T06:59:31+01:00GitHubnoreply@github.com2025-10-17T08:59:31+03:00grammar : use int64_t to avoid int overflows in int schema to grammar conversion logic (#16626)
301ceff6bb253dd306f5404d7ccb3f11fadafe71b521bb4f43380944e94c9a86e305789ba103f5e62bdGittyBursteing0534163997@gmail.com2025-10-17T05:36:40+03:00GitHubnoreply@github.com2025-10-17T10:36:40+08:00SYCL SET operator optimized for F32 tensors (#16350)
3021bb4f43380944e94c9a86e305789ba103f5e62bd683fa6ba4ed3a23b939d3e11e6dc860bd47a0ccfXuan-Son Nguyenson@huggingface.co2025-10-16T19:00:31+02:00GitHubnoreply@github.com2025-10-16T19:00:31+02:00mtmd : support home-cooked Mistral Small Omni (#14928)
303683fa6ba4ed3a23b939d3e11e6dc860bd47a0ccfb22572e97dc51757d3ebe917a5a283385010ec68Pascaladmin@serveurperso.com2025-10-16T16:28:41+02:00GitHubnoreply@github.com2025-10-16T16:28:41+02:00fix: added a normalization step for MathJax-style \[\] and \(\) delimiters (#16599)
304b22572e97dc51757d3ebe917a5a283385010ec687a50cf388a530127cbbdd2a507ef81a451c9d819GittyBursteing0534163997@gmail.com2025-10-16T16:26:21+03:00GitHubnoreply@github.com2025-10-16T15:26:21+02:00sycl : add ARANGE operator (#16362)
30562425b3a7e8b169a3bbe6ce8c7324513a7fcee8c76a63a2998286c1649de2496bccb7d8a46549895Yunzhe Jiayunzhe@calculet.tech2025-10-16T17:34:39+08:00Yunzhe Jiayunzhe@calculet.tech2025-10-16T17:34:39+08:00kv: let seq_rm free whole seq since current hw-op cannot support common-prefix
3067a50cf388a530127cbbdd2a507ef81a451c9d8196f5d924637a15abedb111cbbffd7da5f31c81855Chenguang Li757486878@qq.com2025-10-16T16:41:11+08:00GitHubnoreply@github.com2025-10-16T16:41:11+08:00CANN: format code using .clang-format (#15863)
30776a63a2998286c1649de2496bccb7d8a46549895b67217078aa748b06e1b2269d88dd18c0aca2c26Yunzhe Jiayunzhe@calculet.tech2025-10-16T16:31:15+08:00Yunzhe Jiayunzhe@calculet.tech2025-10-16T16:31:15+08:00kv: fix common-prefix bug by implementing llama_memory_seq_rm
3086f5d924637a15abedb111cbbffd7da5f31c81855adc9b60f190c1016a09f439862fa1cbb302262actakasurazeemtakasurazeem@gmail.com2025-10-16T01:11:33-04:00GitHubnoreply@github.com2025-10-16T08:11:33+03:00common : Update the docs on -t --threads (#16236)
309adc9b60f190c1016a09f439862fa1cbb302262acee50ee1eadff58777ae746827b04de7ba0befc55takuya kodamaa.s.takuya1026@gmail.com2025-10-16T13:10:32+08:00GitHubnoreply@github.com2025-10-16T08:10:32+03:00ggml-cpu: replace putenv with setenv for const-correctness (#16573)
310ee50ee1eadff58777ae746827b04de7ba0befc557adc79c03234de9a20661fd6dbf2d02c32ca7acbyael-works106673277+yael-works@users.noreply.github.com2025-10-16T07:21:28+03:00GitHubnoreply@github.com2025-10-16T12:21:28+08:00SYCL: Add GGML_OP_MEAN operator support (#16009)
3117adc79c03234de9a20661fd6dbf2d02c32ca7acb466c1911ab736f0b7366127edee99f8ee5687417Aleksei Nikiforov103434461+AlekseiNikiforovIBM@users.noreply.github.com2025-10-15T22:43:08+02:00GitHubnoreply@github.com2025-10-15T22:43:08+02:00gguf-py : add support for endian conversion of BF16 data (#16594)
312466c1911ab736f0b7366127edee99f8ee56874170cb7a0683b0529172472d74d21f05470a607f297safranowithbsh155762@gmail.com2025-10-15T22:24:51+03:00GitHubnoreply@github.com2025-10-15T21:24:51+02:00cpu : add FLOOR, CEIL, ROUND and TRUNC unary operators (#16083)
3130cb7a0683b0529172472d74d21f05470a607f297d93f8439b08c4f35e13a41a7366901fdbe770fc8lhezlih@qti.qualcomm.com2025-10-15T10:51:04-07:00GitHubnoreply@github.com2025-10-15T10:51:04-07:00opencl: add q8_0 mm support (#16469)
314d93f8439b08c4f35e13a41a7366901fdbe770fc8f9fb33f2630b4b4ba9081ce9c0c921f8cd8ba4eblhezlih@qti.qualcomm.com2025-10-15T10:48:28-07:00GitHubnoreply@github.com2025-10-15T10:48:28-07:00opencl: fix FA for f32 (#16584)
315f9fb33f2630b4b4ba9081ce9c0c921f8cd8ba4ebf4ce81c45e7bd910e36bf44c253fc5255c49b1e4Aleksander Grygieraleksander.grygier@gmail.com2025-10-15T16:22:20+02:00GitHubnoreply@github.com2025-10-15T16:22:20+02:00Add server-driven parameter defaults and syncing (#16515)
316f4ce81c45e7bd910e36bf44c253fc5255c49b1e417304cbcc1dd24de7741cbe57925d58e90a98ac1Sam/Samuel57896620+cern1710@users.noreply.github.com2025-10-15T23:05:56+09:00GitHubnoreply@github.com2025-10-15T17:05:56+03:00metal: optimise `GGML_OP_SUM` (#16559)
31717304cbcc1dd24de7741cbe57925d58e90a98ac13e3cb19f6449f9168a128eaeae01f8f41b049accGeorgi Gerganovggerganov@gmail.com2025-10-15T16:53:12+03:00GitHubnoreply@github.com2025-10-15T16:53:12+03:00server : fix img token logs (#16595)
3183e3cb19f6449f9168a128eaeae01f8f41b049acc5acd455460f457942d8dd02e3dd9b1eebfce99feXuan-Son Nguyenson@huggingface.co2025-10-15T14:48:08+02:00GitHubnoreply@github.com2025-10-15T14:48:08+02:00llama-quant: add support for mmproj (#16592)
3195acd455460f457942d8dd02e3dd9b1eebfce99fe554fd578a5ed78a10f371f5850c6a69aa83df15aJulius Tischbeinju.tischbein@gmail.com2025-10-15T13:54:15+02:00GitHubnoreply@github.com2025-10-15T14:54:15+03:00CUDA: Changing the CUDA scheduling strategy to spin (#16585)
320554fd578a5ed78a10f371f5850c6a69aa83df15afa882fd2b1bcb663de23af06fdc391489d05b007Georgi Gerganovggerganov@gmail.com2025-10-15T12:51:27+03:00GitHubnoreply@github.com2025-10-15T11:51:27+02:00server : fix mtmd checkpoints (#16591)
321fa882fd2b1bcb663de23af06fdc391489d05b007ffa059034c1e41f3e58363ef3c46d2fcf854bc4dGeorgi Gerganovggerganov@gmail.com2025-10-14T20:33:05+03:00GitHubnoreply@github.com2025-10-14T20:33:05+03:00metal : avoid using Metal's gpuAddress property (#16576)
322ffa059034c1e41f3e58363ef3c46d2fcf854bc4d120bf7046d85a893f064c12abe58bfeebd735f84SavicStefan50296686+SavicStefan@users.noreply.github.com2025-10-14T19:18:05+02:00GitHubnoreply@github.com2025-10-14T19:18:05+02:00vulkan: Add ACC_TYPE_VEC2 implementation (#16203)
323120bf7046d85a893f064c12abe58bfeebd735f844258e0cfe72c72ad2787be1e9f93253e89219455Aman Guptaamangupta052@gmail.com2025-10-14T22:48:08+08:00GitHubnoreply@github.com2025-10-14T07:48:08-07:00CUDA + openCL: fix bug in accessing rms_norm->src while doing fusion (#16577)
3244258e0cfe72c72ad2787be1e9f93253e892194557ea15bb64c81e3813eb0babf9a57e1bc5697f569Jeff Bolzjbolz@nvidia.com2025-10-14T08:53:37-05:00GitHubnoreply@github.com2025-10-14T15:53:37+02:00vulkan: Support FA with K/V in F32 (#16543)
3257ea15bb64c81e3813eb0babf9a57e1bc5697f5699c7185dd28416cf67f5e3b268381f311b5e3da56Jeff Bolzjbolz@nvidia.com2025-10-14T07:51:36-05:00GitHubnoreply@github.com2025-10-14T14:51:36+02:00vulkan: Improve build time for MSVC (#16545)
3269c7185dd28416cf67f5e3b268381f311b5e3da561ee9d0b415cdf5240418c110a18b419f4002b154Johannes Gäßlerjohannesg@5d6.de2025-10-14T14:22:47+02:00GitHubnoreply@github.com2025-10-14T14:22:47+02:00CUDA: enable FA for FP32 KV cache (#16546)
3271ee9d0b415cdf5240418c110a18b419f4002b15448e2fa9fb7c2de1e53808fdb65ec33f916020fc4Aman Guptaamangupta052@gmail.com2025-10-14T19:16:21+08:00GitHubnoreply@github.com2025-10-14T13:16:21+02:00CUDA: use fastdiv + ggml_cuda_mad for mmvf (#16557)
32848e2fa9fb7c2de1e53808fdb65ec33f916020fc45b6913c47b6bc71a6f927805a45387d5657d8b89Aman Guptaamangupta052@gmail.com2025-10-14T19:15:15+08:00GitHubnoreply@github.com2025-10-14T13:15:15+02:00CUDA: add fp kernel for larger batch size MoE (#16512)
3295b6913c47b6bc71a6f927805a45387d5657d8b89bc07349a7f87ba6eb31ed4b0ea9d9a7352185213Anav Prasadanavp@nvidia.com2025-10-14T09:53:49ZGitHubnoreply@github.com2025-10-14T11:53:49+02:00cuda : remove legacy copy-op pointer indirection code (#16485)
330bc07349a7f87ba6eb31ed4b0ea9d9a7352185213e60f241eacec42d3bd7c9edd37d236ebf35132a8Georgi Gerganovggerganov@gmail.com2025-10-14T08:48:50+03:00GitHubnoreply@github.com2025-10-14T08:48:50+03:00server : dynamic token limit for prompt cache (#16560)
331e60f241eacec42d3bd7c9edd37d236ebf35132a8e38b7c6e9e4453e3b3e96d76e38bc2ccb6bce458Georgi Gerganovggerganov@gmail.com2025-10-13T23:07:57+03:00GitHubnoreply@github.com2025-10-13T23:07:57+03:00metal : FA support F32 K and V and head size = 32 (#16531)
332e38b7c6e9e4453e3b3e96d76e38bc2ccb6bce4585016b7286240d29f8f640039989b84ea3a854344Georgi Gerganovggerganov@gmail.com2025-10-13T22:42:37+03:00GitHubnoreply@github.com2025-10-13T22:42:37+03:00graph : support cacheless embeddings with FA and iSWA (#16528)
3335016b7286240d29f8f640039989b84ea3a8543447049736b2dd9011bf819e298b844ebbc4b5afdc9lhezlih@qti.qualcomm.com2025-10-13T11:50:37-07:00GitHubnoreply@github.com2025-10-13T11:50:37-07:00opencl: fix build targeting CL 2 (#16554)
3347049736b2dd9011bf819e298b844ebbc4b5afdc901d2bdc2bc61b676706830305b286b08b9885a41Johannes Gäßlerjohannesg@5d6.de2025-10-13T16:29:45+02:00GitHubnoreply@github.com2025-10-13T17:29:45+03:00CUDA: fix numerical issues in tile FA kernel (#16540)
33501d2bdc2bc61b676706830305b286b08b9885a4156fc38b9655fbe1869d8bd6cfb269418196cea69Jie Fu (傅杰)jiefu@tencent.com2025-10-13T20:48:47+08:00GitHubnoreply@github.com2025-10-13T15:48:47+03:00ggml : fix build broken with -march=armv9-a on MacOS (#16520)
33656fc38b9655fbe1869d8bd6cfb269418196cea691fb9504eb744969a990bfe4cfcf1d3d7a479541cChenguang Li757486878@qq.com2025-10-13T17:01:24+08:00GitHubnoreply@github.com2025-10-13T17:01:24+08:00CANN: fix CPU memory leak in CANN backend (#16549)
3371fb9504eb744969a990bfe4cfcf1d3d7a479541c3f750f8d760ab5a61491e6a9409072dfeee4b4d7Pascaladmin@serveurperso.com2025-10-13T10:55:32+02:00GitHubnoreply@github.com2025-10-13T10:55:32+02:00fix: add remark plugin to render raw HTML as literal text (#16505)
338b67217078aa748b06e1b2269d88dd18c0aca2c2698c8269abefdfa8fde47dfb2769ceebc8b704e86Yunzhe Jiayunzhe@calculet.tech2025-10-13T11:48:14+08:00Yunzhe Jiayunzhe@calculet.tech2025-10-13T16:52:48+08:00add different path for unified and seperate compilation
3393f750f8d760ab5a61491e6a9409072dfeee4b4d7c515fc577166042234241c6bd0da9b08dcbe2bb9Sam/Samuel57896620+cern1710@users.noreply.github.com2025-10-13T16:25:02+08:00GitHubnoreply@github.com2025-10-13T11:25:02+03:00metal: add support for opt_step_sgd (#16539)
340c515fc577166042234241c6bd0da9b08dcbe2bb9f9bc66c3ebcfddb5f09e4b21253623caeb8e414aGeorgi Gerganovggerganov@gmail.com2025-10-13T11:22:27+03:00GitHubnoreply@github.com2025-10-13T11:22:27+03:00ggml : fix scalar path for computing norm (#16558)
341f9bc66c3ebcfddb5f09e4b21253623caeb8e414aa31cf36ad946a13b3a646bf0dadf2a481e89f944hipuddinghuafengchun@gmail.com2025-10-13T08:52:22+08:00GitHubnoreply@github.com2025-10-13T08:52:22+08:00CANN: Update several operators to support FP16 data format (#16251)
342a31cf36ad946a13b3a646bf0dadf2a481e89f94481d54bbfd599811b354c39f04550888168be7780Sam/Samuel57896620+cern1710@users.noreply.github.com2025-10-13T02:43:14+08:00GitHubnoreply@github.com2025-10-12T21:43:14+03:00metal : add opt_step_adamw and op_sum (#16529)
34381d54bbfd599811b354c39f04550888168be7780c7be9febcbafa9af7d1b9443f86475c59c9c5f87Pascaladmin@serveurperso.com2025-10-12T18:06:41+02:00GitHubnoreply@github.com2025-10-12T18:06:41+02:00webui: remove client-side context pre-check and rely on backend for limits (#16506)
344c7be9febcbafa9af7d1b9443f86475c59c9c5f878415f61e23d04427cd0d912fbb9d33b85f849456Neo Zhang Jianyujianyu.zhang@intel.com2025-10-12T21:53:35+08:00GitHubnoreply@github.com2025-10-12T21:53:35+08:00[SYCL] fix UT fault cases: count-equal, argsort, pad OPs (#16521)
3458415f61e23d04427cd0d912fbb9d33b85f8494562c301e91abb92d03c1a682b4b540ba835562a74bMathieu Baudiermbaudier@argeo.org2025-10-12T15:48:03+02:00GitHubnoreply@github.com2025-10-12T15:48:03+02:00ci : add Vulkan on Ubuntu with default packages build (#16532)
3462c301e91abb92d03c1a682b4b540ba835562a74b4b2dae383df708e2afc49c4859a81cd074f5ac10Aldehir Rojashello@alde.dev2025-10-12T08:18:47-05:00GitHubnoreply@github.com2025-10-12T16:18:47+03:00common : handle unicode during partial json parsing (#16526)
3474b2dae383df708e2afc49c4859a81cd074f5ac1041aac5c69b5fb281bc1f486afb053f78101bb39eGeorgi Gerganovggerganov@gmail.com2025-10-12T09:29:13+03:00GitHubnoreply@github.com2025-10-12T09:29:13+03:00common : update presets (#16504)
34841aac5c69b5fb281bc1f486afb053f78101bb39ea2fba89a426ff8005d303c73f0436e7e67368b70sirus20x6sirus20x6@users.noreply.github.com2025-10-12T00:25:37-05:00GitHubnoreply@github.com2025-10-12T08:25:37+03:00ggml : Fix FP16 ELU positive branch (#16519)
349a2fba89a426ff8005d303c73f0436e7e67368b7020cc625edc2264aae2779e71bef1593e6a4e8c43Daniel Beveniusdaniel.bevenius@gmail.com2025-10-12T07:19:06+02:00GitHubnoreply@github.com2025-10-12T07:19:06+02:00hparams : add check for layer index in is_recurrent (#16511)
35020cc625edc2264aae2779e71bef1593e6a4e8c4311f0af5504252e453d57406a935480c909e3ff37sirus20x6sirus20x6@users.noreply.github.com2025-10-12T00:15:00-05:00GitHubnoreply@github.com2025-10-12T08:15:00+03:00ggml: Correct SVE implementation in ggml_vec_dot_f16_unroll (#16518)
35111f0af5504252e453d57406a935480c909e3ff37a3cb04744fb5c591985f53b749fef5407d07a145Johannes Gäßlerjohannesg@5d6.de2025-10-11T20:54:32+02:00GitHubnoreply@github.com2025-10-11T20:54:32+02:00CUDA: faster tile FA, add oob checks, more HSs (#16492)
352a3cb04744fb5c591985f53b749fef5407d07a1454a8fbe0a5eb75f339782ebcf29c17848122184d3Georgi Gerganovggerganov@gmail.com2025-10-11T16:54:10+03:00GitHubnoreply@github.com2025-10-11T16:54:10+03:00metal : fix mul-mm condition + fix mul-mv permuted kernels (#16494)
3534a8fbe0a5eb75f339782ebcf29c17848122184d331d0ff1869aa2ea31f4e96d5877d0343e9a2171bPascaladmin@serveurperso.com2025-10-11T15:50:49+02:00GitHubnoreply@github.com2025-10-11T15:50:49+02:00feat: render user content as markdown option (#16358)
35431d0ff1869aa2ea31f4e96d5877d0343e9a2171b97870e64975b26c5e06a3540a8dc0ff601351e86Yann Follet131855179+YannFollet@users.noreply.github.com2025-10-11T21:39:04+08:00GitHubnoreply@github.com2025-10-11T16:39:04+03:00server / ranking : add sorting and management of top_n (#16403)
35597870e64975b26c5e06a3540a8dc0ff601351e86477a66b03501cf3bd067f8968b77ca4d053ff1bdDiego Devesaslarengh@gmail.com2025-10-11T04:02:26-07:00GitHubnoreply@github.com2025-10-11T13:02:26+02:00cuda : avoid initializing unused devices (#16510)
356477a66b03501cf3bd067f8968b77ca4d053ff1bde60f01d941bc5b7fae62dd57fee4cec76ec0ea6eamirai2189905406+amirai21@users.noreply.github.com2025-10-11T11:33:41+03:00GitHubnoreply@github.com2025-10-11T10:33:41+02:00convert : correctly handle LLaMA tokenizer for Jamba (#16470)
357e60f01d941bc5b7fae62dd57fee4cec76ec0ea6e81086cd6a3ca1252f0dc0f938171648399179c53Georgi Gerganovggerganov@gmail.com2025-10-10T22:15:05+03:00GitHubnoreply@github.com2025-10-10T22:15:05+03:00server : fix division by zero when reporting stats (#16501)
35881086cd6a3ca1252f0dc0f938171648399179c5368ee98ae181a5c83a5cc6261daeee69a1f588c15Georgi Gerganovggerganov@gmail.com2025-10-10T17:17:31+03:00GitHubnoreply@github.com2025-10-10T17:17:31+03:00vocab : mark EOT token for Granite models (#16499)
35968ee98ae181a5c83a5cc6261daeee69a1f588c15cdb6da468cc33323955a523738d2e1675aeb5e9aRadoslav Gerganovrgerganov@gmail.com2025-10-10T17:11:07+03:00GitHubnoreply@github.com2025-10-10T16:11:07+02:00server : return HTTP 400 if prompt exceeds context length (#16486)
360cdb6da468cc33323955a523738d2e1675aeb5e9a6d69ab3f262eca3f0c6ad0ce075d2c80b9924d0eRadoslav Gerganovrgerganov@gmail.com2025-10-10T13:22:27+03:00GitHubnoreply@github.com2025-10-10T13:22:27+03:00server : log requests to /v1/completions (#16495)
3616d69ab3f262eca3f0c6ad0ce075d2c80b9924d0e1faa13a1187051af66b0fd9f0d6effe4c77f0b3ePrajwal B Mehendarkarprajwal.b.mehendarkar@ibm.com2025-10-10T13:45:46+05:30GitHubnoreply@github.com2025-10-10T11:15:46+03:00cmake : Dont define XOPENSOURCE on AIX (#16481)
3621faa13a1187051af66b0fd9f0d6effe4c77f0b3e1deee0f8d494981c32597dca8b5f8696d399b0f2Pascaladmin@serveurperso.com2025-10-09T22:54:57+02:00GitHubnoreply@github.com2025-10-09T22:54:57+02:00webui: updated the chat service to only include max_tokens in the req… (#16489)
3631deee0f8d494981c32597dca8b5f8696d399b0f2d00cbea63c671cd85a57adaa50abf60b3b87d86fdudutasimona.gherman@gmail.com2025-10-09T22:11:15+03:00GitHubnoreply@github.com2025-10-09T21:11:15+02:00cpu : optimize the ggml NORM operation (#15953)
364d00cbea63c671cd85a57adaa50abf60b3b87d86f8328fd4bae76fc027f8ca0e9a05acd3788dabe3fGeorgi Gerganovggerganov@gmail.com2025-10-09T18:54:51+03:00GitHubnoreply@github.com2025-10-09T18:54:51+03:00server : host-memory prompt caching (#16391)
3658328fd4bae76fc027f8ca0e9a05acd3788dabe3f56b4795842d852152222ca7a2d4304008facf1b9Pascaladmin@serveurperso.com2025-10-09T17:36:29+02:00GitHubnoreply@github.com2025-10-09T17:36:29+02:00No markdown in cot (#16483)
36656b4795842d852152222ca7a2d4304008facf1b92c0d875ae6c6043fbbafd2831e160955e9ca6af1Daniel Beveniusdaniel.bevenius@gmail.com2025-10-09T14:35:22+02:00GitHubnoreply@github.com2025-10-09T14:35:22+02:00model-conversion : add support for SentenceTransformers (#16387)
3672c0d875ae6c6043fbbafd2831e160955e9ca6af1aa4711d369b0d4adb6802b98ad6ea362f838710esudhiarmsudhi.sathyavathy@arm.com2025-10-09T09:13:18+01:00GitHubnoreply@github.com2025-10-09T10:13:18+02:00ci: add ARM64 Kleidiai build and test support (#16462)
368aa4711d369b0d4adb6802b98ad6ea362f838710ed80d6d2400b8faa80654c723cb5bdf9fc8f4db06Chenguang Li757486878@qq.com2025-10-09T15:50:25+08:00GitHubnoreply@github.com2025-10-09T15:50:25+08:00CANN: Improve ACL graph matching (#16166)
369d80d6d2400b8faa80654c723cb5bdf9fc8f4db06b2602137557b2b28a39e03612717d85ead9a6f5aCharles Xucharles.xu@arm.com2025-10-09T09:29:17+02:00GitHubnoreply@github.com2025-10-09T10:29:17+03:00kleidiai: kernel interface refactoring (#16460)
370b2602137557b2b28a39e03612717d85ead9a6f5ae08db4259521de493b7aeb49dadf29ebd1ee966aNeo Zhang Jianyujianyu.zhang@intel.com2025-10-09T15:25:11+08:00GitHubnoreply@github.com2025-10-09T10:25:11+03:00[SYCL] refactor soft_max, add soft_max_back (#16472)
371e08db4259521de493b7aeb49dadf29ebd1ee966a12bbc3fa50b6df03318a4451c9a2210200a0b28dSaba Fallah10401143+sfallah@users.noreply.github.com2025-10-09T08:39:18+02:00GitHubnoreply@github.com2025-10-09T09:39:18+03:00model: EmbeddingGemma Adding Support for SentenceTransformers Dense Modules (#16367)
37212bbc3fa50b6df03318a4451c9a2210200a0b28d9d0882840e6c3fb62965d03af0e22880ea90e012Pascaladmin@serveurperso.com2025-10-08T22:18:41+02:00GitHubnoreply@github.com2025-10-08T23:18:41+03:00refactor: centralize CoT parsing in backend for streaming mode (#16394)
3739d0882840e6c3fb62965d03af0e22880ea90e012d2ee056e1df0fdf646270fb5621e9f92084b59a7ai-fonsilength-amiss-7k@icloud.com2025-10-08T20:21:46+02:00GitHubnoreply@github.com2025-10-08T20:21:46+02:00Disable CUDA host buffers on integrated GPUs (#16308)
374d2ee056e1df0fdf646270fb5621e9f92084b59a7b2c08c9ec4cd59ed88c1ebdd94f109af5ccc978eissixx46835150+issixx@users.noreply.github.com2025-10-08T17:20:18+09:00GitHubnoreply@github.com2025-10-08T11:20:18+03:00server : fix cancel pending task (#16467)
375b2c08c9ec4cd59ed88c1ebdd94f109af5ccc978e7fdd16b432d247121fe5fe7b21f8805f85266c85Georgi Gerganovggerganov@gmail.com2025-10-08T10:57:53+03:00GitHubnoreply@github.com2025-10-08T10:57:53+03:00metal : mark FA blocks (#16372)
3767fdd16b432d247121fe5fe7b21f8805f85266c8574b8fc17f92ada295a648e3c5eb28f46bca7d892Georgi Gerganovggerganov@gmail.com2025-10-08T10:57:29+03:00GitHubnoreply@github.com2025-10-08T10:57:29+03:00server : improve context checkpoint logic (#16440)
37774b8fc17f92ada295a648e3c5eb28f46bca7d892aeaf8a36f06b5810f5ae4bbefe26edb33925cf5eReese Levinereeselevine1@gmail.com2025-10-07T13:48:56-07:00GitHubnoreply@github.com2025-10-07T13:48:56-07:00ggml webgpu: profiling, CI updates, reworking of command submission (#16452)
378aeaf8a36f06b5810f5ae4bbefe26edb33925cf5edf1b612e29ba97a2e67db339b1e8c7465702b7e8Tarek Dakhrantarek@liquid.ai2025-10-07T20:03:35+02:00GitHubnoreply@github.com2025-10-07T20:03:35+02:00llama : support LiquidAI LFM2-MoE hybrid model (#16464)
379df1b612e29ba97a2e67db339b1e8c7465702b7e84e0388aa8a1f4ef1065701dc9c3947aea0a1b9a5Georgi Gerganovggerganov@gmail.com2025-10-07T15:57:14+03:00GitHubnoreply@github.com2025-10-07T15:57:14+03:00server : add `/v1/health` endpoint (#16461)
3804e0388aa8a1f4ef1065701dc9c3947aea0a1b9a5ef4c5b87ea2556ff8ca99cca3abdf48bdbca22f2Sascha Rogmann59577610+srogmann@users.noreply.github.com2025-10-07T11:11:08+02:00GitHubnoreply@github.com2025-10-07T11:11:08+02:00webui : added download action (#13552) (#16282)
381ef4c5b87ea2556ff8ca99cca3abdf48bdbca22f2c61ae20d05bd4fdd8551311325a2845336449426Georgi Gerganovggerganov@gmail.com2025-10-07T10:32:32+03:00GitHubnoreply@github.com2025-10-07T10:32:32+03:00presets : fix pooling param for embedding models (#16455)
382c61ae20d05bd4fdd8551311325a28453364494260123ff38f53d34752f29239a29d0e40a6dc4110fRadoslav Gerganovrgerganov@gmail.com2025-10-07T09:59:13+03:00GitHubnoreply@github.com2025-10-07T06:59:13Zrpc : update documentation (#16441)
3830123ff38f53d34752f29239a29d0e40a6dc4110f0a319bb75ed29d968e2a9b544011b09ccb932915Georgi Gerganovggerganov@gmail.com2025-10-07T08:24:17+03:00GitHubnoreply@github.com2025-10-07T08:24:17+03:00memory : use sequential equal splits for recurrent modules (#16442)
3840a319bb75ed29d968e2a9b544011b09ccb9329151d6092fc72f4d10f4486ac95edfd414bc08b62b8Georgi Gerganovggerganov@gmail.com2025-10-07T08:23:30+03:00GitHubnoreply@github.com2025-10-07T08:23:30+03:00metal : add support for non-padded FA KV (#16148)
3851d6092fc72f4d10f4486ac95edfd414bc08b62b88ae32dc9ecd7aeebf3a5b43557e0552a0a04cd4fGeorgi Gerganovggerganov@gmail.com2025-10-07T08:22:35+03:00GitHubnoreply@github.com2025-10-07T08:22:35+03:00tests : add -INF blocks to the KQ mask in the FA tests (#16380)
3868ae32dc9ecd7aeebf3a5b43557e0552a0a04cd4f3df2244df40c67dfd6ad548b40ccc507a066af2bGeorgi Gerganovggerganov@gmail.com2025-10-07T08:21:40+03:00GitHubnoreply@github.com2025-10-07T08:21:40+03:00metal : various optimizations + refactoring (#16446)
3873df2244df40c67dfd6ad548b40ccc507a066af2bc08002a1988348403a5fd59b1fa3de3a10a6f92fGadflyii34758915+Gadflyii@users.noreply.github.com2025-10-06T12:55:53-05:00GitHubnoreply@github.com2025-10-06T19:55:53+02:00llama : add --no-host to disable host buffers (#16310)
388c08002a1988348403a5fd59b1fa3de3a10a6f92f3a002afafa8e06555f11aed88b02d055d8a166f3Gabe Goodhartghart@us.ibm.com2025-10-06T10:59:40-06:00GitHubnoreply@github.com2025-10-06T18:59:40+02:00chat : Granite Docling stopping (#16438)
3893a002afafa8e06555f11aed88b02d055d8a166f3a23b9bdbd3b64ce172f9962249f432d01aea7437Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-06T17:40:21+02:00GitHubnoreply@github.com2025-10-06T17:40:21+02:00ci : refactor sdk caching to minimize storage (#16414)
390a23b9bdbd3b64ce172f9962249f432d01aea743704e632a4aab8e6bfff0f8bc216b36ceb1e199ff9Georgi Gerganovggerganov@gmail.com2025-10-06T16:05:27+03:00GitHubnoreply@github.com2025-10-06T16:05:27+03:00ggml : fix unaligned access in AMX code (#16315)
39104e632a4aab8e6bfff0f8bc216b36ceb1e199ff9a80ff183abe4e5a76316257ffa597da41b3b6fa0Daniel Beveniusdaniel.bevenius@gmail.com2025-10-06T14:56:59+02:00GitHubnoreply@github.com2025-10-06T14:56:59+02:00ci : remove missing reranker model files (#16444)
392a80ff183abe4e5a76316257ffa597da41b3b6fa01d49ca37594fb49db6aa9518ba7c512e5ccd0108Daniel Beveniusdaniel.bevenius@gmail.com2025-10-06T14:17:12+02:00GitHubnoreply@github.com2025-10-06T14:17:12+02:00ggml-cpu : fix leftover handling in ggml_vec_scale_f32 for SVE (#16443)
3931d49ca37594fb49db6aa9518ba7c512e5ccd0108c5fef0fcea3b978a2318b1af170209ecec7c37b4Yuannanyuannan@users.noreply.github.com2025-10-06T09:29:56ZGitHubnoreply@github.com2025-10-06T12:29:56+03:00nix : removed metal for nix (#16118)
394c5fef0fcea3b978a2318b1af170209ecec7c37b4ca71fb9b368e3db96e028f80c4c9df6b6b370eddOleksandr Kuvshynov661042+okuvshynov@users.noreply.github.com2025-10-06T03:53:31-04:00GitHubnoreply@github.com2025-10-06T10:53:31+03:00server: update readme to mention n_past_max metric (#16436)
395ca71fb9b368e3db96e028f80c4c9df6b6b370edd35266573b968e1c947b367782fb4b3eddbb4f3c0Gabe Goodhartghart@us.ibm.com2025-10-05T06:57:47-06:00GitHubnoreply@github.com2025-10-05T14:57:47+02:00model : Granite docling + Idefics3 preprocessing (SmolVLM) (#16206)
39635266573b968e1c947b367782fb4b3eddbb4f3c086df2c9ae4f2f1ee63d2558a9dc797b98524639bReese Levinereeselevine1@gmail.com2025-10-04T20:59:31-07:00GitHubnoreply@github.com2025-10-04T20:59:31-07:00ggml webgpu: actually add softmax, fix rms_norm offset (#16400)
39786df2c9ae4f2f1ee63d2558a9dc797b98524639bf39283960b58a92ecc0c72567711318b20e22b55Eve139727413+netrunnereve@users.noreply.github.com2025-10-04T20:04:27ZGitHubnoreply@github.com2025-10-04T22:04:27+02:00vulkan: use a more appropriate amount of threads when generating shaders (#16418)
398f39283960b58a92ecc0c72567711318b20e22b55898acba6816ad23b6a9491347d30e7570bffadfdRadoslav Gerganovrgerganov@gmail.com2025-10-04T16:22:45+03:00GitHubnoreply@github.com2025-10-04T16:22:45+03:00rpc : check src buffer when copying tensor (#16421)
399898acba6816ad23b6a9491347d30e7570bffadfde29acf74fea996014380d59d31aa504ae8964258Radoslav Gerganovrgerganov@gmail.com2025-10-04T12:49:16+03:00GitHubnoreply@github.com2025-10-04T12:49:16+03:00rpc : add support for multiple devices (#16276)
400e29acf74fea996014380d59d31aa504ae8964258128d522c04286e019666bd6ee4d18e3fbf8772e2Aclyaclysia@gmail.com2025-10-04T11:42:56+02:00GitHubnoreply@github.com2025-10-04T11:42:56+02:00vulkan : incremental shader builds (#16341)
401128d522c04286e019666bd6ee4d18e3fbf8772e2f6dcda390004b627ef30af378d0c01ad2519289ePascaladmin@serveurperso.com2025-10-03T20:51:48+02:00GitHubnoreply@github.com2025-10-03T21:51:48+03:00chat : support Magistral thinking (#16413)
402f6dcda390004b627ef30af378d0c01ad2519289e606a73f53175077429484b23dcf799f69a31d0bdddh0dylanhalladay02@icloud.com2025-10-03T13:34:51-05:00GitHubnoreply@github.com2025-10-03T21:34:51+03:00server : context checkpointing for hybrid and recurrent models (#16382)
403606a73f53175077429484b23dcf799f69a31d0bd946f71ed9ade07e319859b5ce656144140e066fbGeorgi Gerganovggerganov@gmail.com2025-10-03T19:18:56+03:00GitHubnoreply@github.com2025-10-03T19:18:56+03:00metal : fix loop bound in ggml_mem_ranges (#16412)
404946f71ed9ade07e319859b5ce656144140e066fb638d330246954e88dffc22ce01fec15e6894e544Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-03T14:40:25+02:00GitHubnoreply@github.com2025-10-03T14:40:25+02:00llama : fix shapes for bert/mpt q/k norm (#16409)
405638d330246954e88dffc22ce01fec15e6894e54484c8e305e8010a1a3d43bdd0a25f737ac67809a4Aclyaclysia@gmail.com2025-10-03T13:49:08+02:00GitHubnoreply@github.com2025-10-03T13:49:08+02:00ggml : fix graph reallocation with multiple chunks (#16396)
40684c8e305e8010a1a3d43bdd0a25f737ac67809a42aaf0a2a2056d75d0dd53ab8a181473760e6ab22Aleksander Grygieraleksander.grygier@gmail.com2025-10-03T12:51:40+02:00GitHubnoreply@github.com2025-10-03T12:51:40+02:00Fix missing messages on sibling navigation (#16408)
4072aaf0a2a2056d75d0dd53ab8a181473760e6ab220e1f83855609d73beaf05d818640b6cfd39d287bJeff Bolzjbolz@nvidia.com2025-10-03T05:50:46-05:00GitHubnoreply@github.com2025-10-03T12:50:46+02:00vulkan: Replace uses of maxMemoryAllocationSize and VK_WHOLE_SIZE (#16354)
4080e1f83855609d73beaf05d818640b6cfd39d287bad126479c25cf983a0f994a08ba0911cf49ed62bJeff Bolzjbolz@nvidia.com2025-10-03T04:52:46-05:00GitHubnoreply@github.com2025-10-03T11:52:46+02:00vulkan: Fix FA coopmat1 invalid array indexing (#16365)
409ad126479c25cf983a0f994a08ba0911cf49ed62b77233277c912d495d1069543ca540c21a2f9e5b4Daniel Beveniusdaniel.bevenius@gmail.com2025-10-03T11:45:16+02:00GitHubnoreply@github.com2025-10-03T11:45:16+02:00ci : change macos-13 to macos-15-intel (#16401)
41077233277c912d495d1069543ca540c21a2f9e5b4e308efda8e54e3f07578937a45a5d83624eb7970Aleksander Grygieraleksander.grygier@gmail.com2025-10-03T11:30:39+02:00GitHubnoreply@github.com2025-10-03T11:30:39+02:00Capture model name only after first token (streaming) or completed request (#16405)
411e308efda8e54e3f07578937a45a5d83624eb7970136bda78c5679b266362b39c58338fff46fd2592Jeff Bolzjbolz@nvidia.com2025-10-03T03:33:08-05:00GitHubnoreply@github.com2025-10-03T10:33:08+02:00vulkan: in flash attention, bounds check against nem1 (don't rely on GGML_KQ_MASK_PAD) (#16316)
412136bda78c5679b266362b39c58338fff46fd25925113efd34ceda709292de26c72716c49e024fb32Aleksander Grygieraleksander.grygier@gmail.com2025-10-03T09:11:34+02:00GitHubnoreply@github.com2025-10-03T10:11:34+03:00webui : Fix messages payload sent to chat completions (#16402)
4135113efd34ceda709292de26c72716c49e024fb32d64c8104f090b27b1f99e8da5995ffcfa6b726e2Pascaladmin@serveurperso.com2025-10-03T08:01:31+02:00GitHubnoreply@github.com2025-10-03T08:01:31+02:00fix: track viewportHeight via window.innerHeight to avoid unwanted scrolling (#16356)
414d64c8104f090b27b1f99e8da5995ffcfa6b726e2ef07a4090672a3438d7f64f197795d7dc1c18957Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-02T20:10:12+02:00GitHubnoreply@github.com2025-10-02T20:10:12+02:00test-barrier : do not use more threads than physically available (#16389)
415ef07a4090672a3438d7f64f197795d7dc1c1895734fcc5a4ace8c69476ef2ea3857f39a60334acc4Reese Levinereeselevine1@gmail.com2025-10-02T11:00:31-07:00GitHubnoreply@github.com2025-10-02T11:00:31-07:00ggml webgpu: add support for soft_max, optimize rms_norm (#16357)
41634fcc5a4ace8c69476ef2ea3857f39a60334acc491a2a5655658bb9ab77894716b82fae7ecb4b4d1Piotr Wilkin (ilintar)piotr.wilkin@syndatis.com2025-10-02T19:43:22+02:00GitHubnoreply@github.com2025-10-02T20:43:22+03:00model : Apertus model implementation (#15852)
41791a2a5655658bb9ab77894716b82fae7ecb4b4d172ee736c447be4ededb51ab97904b4d33c90c261R0CKSTARyeahdongcn@gmail.com2025-10-02T21:29:56+08:00GitHubnoreply@github.com2025-10-02T16:29:56+03:00musa: update compile flags (#16265)
41872ee736c447be4ededb51ab97904b4d33c90c261f09aefaa84d7f4d5df3f400f67944b94fef5b795Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-02T13:51:36+02:00GitHubnoreply@github.com2025-10-02T13:51:36+02:00ci : fix ubuntu-latest-cmake-rpc (disable ccache) (#16388)
419f09aefaa84d7f4d5df3f400f67944b94fef5b795bbd32bc0384fbfcf07369617de58856b1e0e95a3Eve139727413+netrunnereve@users.noreply.github.com2025-10-02T08:10:07ZGitHubnoreply@github.com2025-10-02T10:10:07+02:00ci: update vulkan ci (#16294)
420bbd32bc0384fbfcf07369617de58856b1e0e95a32be72c2b121ee99f33927149265ce6073ade9e59Georgi Gerganovggerganov@gmail.com2025-10-02T10:35:43+03:00GitHubnoreply@github.com2025-10-02T10:35:43+03:00ci : fix clean-up of old logs (#16381)
4212be72c2b121ee99f33927149265ce6073ade9e5995ce0985449c52fb9d6849b95563f7cf933ea0e3Neo Zhang Jianyujianyu.zhang@intel.com2025-10-02T15:16:25+08:00GitHubnoreply@github.com2025-10-02T10:16:25+03:00SYCL: Update to oneAPI 2025.2 (#16371)
42295ce0985449c52fb9d6849b95563f7cf933ea0e3c8dedc9999eccf7821a9fe5b29f10e8d075e2217uvoscarl@uvos.xyz2025-10-02T05:52:59+02:00GitHubnoreply@github.com2025-10-02T05:52:59+02:00HIP: add IMbackK to codeowner (#16375)
423c8dedc9999eccf7821a9fe5b29f10e8d075e2217e95fec640f43623911a2cd5bda8b19b1898c530cuvoscarl@uvos.xyz2025-10-01T23:32:39+02:00GitHubnoreply@github.com2025-10-01T23:32:39+02:00CI: reenable cdna in rocm docker builds (#16376)
424e95fec640f43623911a2cd5bda8b19b1898c530cded67b94446ef4f7fd988dbde7a12deef9870c13uvoscarl@uvos.xyz2025-10-01T23:09:25+02:00GitHubnoreply@github.com2025-10-01T23:09:25+02:00HIP: Disable ROCWMMA fattn on CDNA when compiled against ROCWMMA 2.0.0 (#16221)
425ded67b94446ef4f7fd988dbde7a12deef9870c131fe4e38cc20af058ed320bd46cac934991190056Shunta Saitoshunta.saito@gmail.com2025-10-02T06:08:15+09:00GitHubnoreply@github.com2025-10-01T23:08:15+02:00llama : parameter conversion and loading fixes for PLaMo2 variants (#16075)
4261fe4e38cc20af058ed320bd46cac9349911900564201deae9c2ae4732db8957b6ce0808d02ec597cuvoscarl@uvos.xyz2025-10-01T20:18:03+02:00GitHubnoreply@github.com2025-10-01T20:18:03+02:00ci: Properly install rocwmma for hip builds (#16305)
4274201deae9c2ae4732db8957b6ce0808d02ec597c764799279f801c696503bacca490b4358587d94cAdrien Gallouëtadrien@gallouet.fr2025-10-01T19:22:18+02:00GitHubnoreply@github.com2025-10-01T20:22:18+03:00common: introduce http.h for httplib-based client (#16373)
428764799279f801c696503bacca490b4358587d94c2a9b63383a448ec18c754dfdc6e95cb853940a52Aleksander Grygieraleksander.grygier@gmail.com2025-10-01T18:18:10+02:00GitHubnoreply@github.com2025-10-01T18:18:10+02:00Conversation action dialogs as singletons from Chat Sidebar + apply conditional rendering for Actions Dropdown for Chat Conversation Items (#16369)
4292a9b63383a448ec18c754dfdc6e95cb853940a521104ca1a1c926b832274694a2773a17066982ad0Aleksander Grygieraleksander.grygier@gmail.com2025-10-01T15:54:42+02:00GitHubnoreply@github.com2025-10-01T15:54:42+02:00Improve code block color theming (#16325)
4301104ca1a1c926b832274694a2773a17066982ad04f1575921cac9b489fe8f8bbf20aff985e7da45eSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-10-01T14:09:52+02:00GitHubnoreply@github.com2025-10-01T14:09:52+02:00ci : use registry cache for docker builds (#16366)
431132d673554e65282e92505f4f77401ab971528bbaa9538a63aef35f0224b2502f13b19d650a8ce04Eve139727413+netrunnereve@users.noreply.github.com2025-10-01T07:56:36ZGitHubnoreply@github.com2025-10-01T09:56:36+02:00vulkan: make ggml_vk_default_dispatcher support older vulkan headers (#16345)
432aa9538a63aef35f0224b2502f13b19d650a8ce04e74c92e84236b2bab3f3c77bee4ead94928be360Aleksander Grygieraleksander.grygier@gmail.com2025-10-01T07:40:26+02:00GitHubnoreply@github.com2025-10-01T08:40:26+03:00webui: Remove running `llama-server` within WebUI `dev.sh` script (#16363)
433e74c92e84236b2bab3f3c77bee4ead94928be360b2ba81dbe07b6dbea9c96b13346c66973dede32cBartowski3266127+bartowski1182@users.noreply.github.com2025-09-30T16:24:36-04:00GitHubnoreply@github.com2025-09-30T22:24:36+02:00model : support GLM 4.6 (make a few NextN/MTP tensors not required) (#16359)
434b2ba81dbe07b6dbea9c96b13346c66973dede32cbf6f3b3a1965d70e07ca94aab7b01268fe483e96Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-30T21:41:42+02:00GitHubnoreply@github.com2025-09-30T21:41:42+02:00ci : fix ccache key for ubuntu-cpu-cmake (#16355)
435bf6f3b3a1965d70e07ca94aab7b01268fe483e967c156df4148eb524b22f7f363bfd2df00f264e0bAdrien Gallouëtangt@huggingface.co2025-09-30T19:52:41+02:00GitHubnoreply@github.com2025-09-30T20:52:41+03:00common : disable progress bar without a tty (#16352)
4367c156df4148eb524b22f7f363bfd2df00f264e0b16b0ca0d2e63fc1a0a43795f8552f9cb61d9f7a5lhezlih@qti.qualcomm.com2025-09-30T10:45:45-07:00GitHubnoreply@github.com2025-09-30T10:45:45-07:00opencl: support pad_ext (#15888)
43716b0ca0d2e63fc1a0a43795f8552f9cb61d9f7a58d78cd2613ccdeb3cc86f59bc8f9ddd31cfbd3edPascaladmin@serveurperso.com2025-09-30T19:18:54+02:00GitHubnoreply@github.com2025-09-30T19:18:54+02:00Chatapi ignore empty sampling (#16330)
4388d78cd2613ccdeb3cc86f59bc8f9ddd31cfbd3edd1c84a662daa91be975863913a59975b11458141Reese Levinereeselevine1@gmail.com2025-09-30T09:57:51-07:00GitHubnoreply@github.com2025-09-30T09:57:51-07:00ggml webgpu: support for rope,div,sub,glu,scale,cont operators (#16187)
439d1c84a662daa91be975863913a59975b11458141364a7a6d4a786e98947c8a90430ea581213c0ba9lhezlih@qti.qualcomm.com2025-09-30T09:55:13-07:00GitHubnoreply@github.com2025-09-30T09:55:13-07:00opencl: support ne3 in get_rows (#15866)
440364a7a6d4a786e98947c8a90430ea581213c0ba92df5bcf357dba0c49b4df1d684b2ffc9e88b7054Adrien Gallouëtangt@huggingface.co2025-09-30T16:39:44+02:00GitHubnoreply@github.com2025-09-30T17:39:44+03:00common : remove common_has_curl() (#16351)
4412df5bcf357dba0c49b4df1d684b2ffc9e88b7054075c01567bb7e93965ae60b7e119182c3e68f3e9Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-30T15:38:01+02:00GitHubnoreply@github.com2025-09-30T15:38:01+02:00ci : disable ccache for android (#16348)
442075c01567bb7e93965ae60b7e119182c3e68f3e9a014310374a16f9204f2bcc1b458fc1eda67e469Georgi Gerganovggerganov@gmail.com2025-09-30T13:42:39+03:00Georgi Gerganovggerganov@gmail.com2025-09-30T13:53:55+03:00ggml : bump version to 0.9.4 (ggml/1363)
44398c8269abefdfa8fde47dfb2769ceebc8b704e862790ecc304e384d9075c55d0628d8bf440025dc4Yunzhe Jiayunzhe@calculet.tech2025-09-30T17:42:57+08:00Yunzhe Jiayunzhe@calculet.tech2025-09-30T17:44:27+08:00server: adapt to calrt
4442790ecc304e384d9075c55d0628d8bf440025dc4f528e97d894b4d93934a0711a00253eea980e799Yunzhe Jiayunzhe@calculet.tech2025-09-30T17:37:58+08:00Yunzhe Jiayunzhe@calculet.tech2025-09-30T17:44:27+08:00adapt to calrt
445f528e97d894b4d93934a0711a00253eea980e799dcbe998a6d8f13b6c9cb36bbe7b355ced2ac3842Yunzhe Jiayunzhe@calculet.tech2025-09-11T14:13:35+08:00Yunzhe Jiayunzhe@calculet.tech2025-09-30T17:44:27+08:00add calrt mtmd support
446a014310374a16f9204f2bcc1b458fc1eda67e46935fb82497ec6c5904b0adb7e1c881a76c1c692dbanavp-nvidiaanavp@nvidia.com2025-09-30T08:13:22ZGitHubnoreply@github.com2025-09-30T11:13:22+03:00cuda : Enable CUDA Graph usage for Nemotron Nano v2 (NemotronH) (#16328)
44735fb82497ec6c5904b0adb7e1c881a76c1c692db3c62aed89fbe3e15f6f82035e805a0a1add26306Georgi Gerganovggerganov@gmail.com2025-09-30T11:03:23+03:00GitHubnoreply@github.com2025-09-30T11:03:23+03:00metal : dynamic simdgroups for MV kernels (#16340)
4483c62aed89fbe3e15f6f82035e805a0a1add26306f1eb1cb1eba042b2d583234e80f65e2e225d8995Adrien Gallouëtangt@huggingface.co2025-09-30T09:36:33+02:00GitHubnoreply@github.com2025-09-30T10:36:33+03:00common : simplify etag tracking by removing json (#16342)
449f1eb1cb1eba042b2d583234e80f65e2e225d8995de41f2b7bffab7582944b213ce12b605f8cf6e0fCharles Xucharles.xu@arm.com2025-09-30T09:07:20+02:00GitHubnoreply@github.com2025-09-30T10:07:20+03:00kleidiai : fix work size and threads sync for fp16 (#16246)
450de41f2b7bffab7582944b213ce12b605f8cf6e0fa74a0d69f34f52fa10d4f0a7ce749fb3490d0774lhezlih@qti.qualcomm.com2025-09-29T22:30:16-07:00GitHubnoreply@github.com2025-09-30T08:30:16+03:00codeowners: add codeowners for opencl backend (#16344)
451a74a0d69f34f52fa10d4f0a7ce749fb3490d07745f7e166cbf7b9ca928c7fad990098ef32358ac75Jeff Bolzjbolz@nvidia.com2025-09-29T19:26:34-05:00GitHubnoreply@github.com2025-09-29T19:26:34-05:00tests: override test_set_rows::max_nmse_err to allow for occasional rounding differences (#16295)
4525f7e166cbf7b9ca928c7fad990098ef32358ac75d72f5f7ba260b546190338b0b76f2f152581424fPascaladmin@serveurperso.com2025-09-29T18:49:47+02:00GitHubnoreply@github.com2025-09-29T18:49:47+02:00Fix thinking blocks with quotes + add handling `[THINK]...[/THINK]` blocks (#16326)
453d72f5f7ba260b546190338b0b76f2f152581424fb77e6c18e1a6fac5705ed95f03af5436d67484c1Georgi Gerganovggerganov@gmail.com2025-09-29T17:51:48+03:00GitHubnoreply@github.com2025-09-29T17:51:48+03:00ci : add AMD runners and workflows (#16249)
454b77e6c18e1a6fac5705ed95f03af5436d67484c12ddd3f2356c313385fafc19a9f0ce678c8fe03eealex-spacemitjinghui.huang@spacemit.com2025-09-29T22:50:44+08:00GitHubnoreply@github.com2025-09-29T17:50:44+03:00ggml: riscv: add riscv spacemit backend (#15288)
4552ddd3f2356c313385fafc19a9f0ce678c8fe03ee4d3d455d3c8719eb8f206de328d1b9ba191efd7cGeorgi Gerganovggerganov@gmail.com2025-09-29T16:50:52+03:00Georgi Gerganovggerganov@gmail.com2025-09-29T17:43:58+03:00sync : ggml
4564d3d455d3c8719eb8f206de328d1b9ba191efd7cc9b1c06467aca8d30fafcab51292bcb285df5b0dGeorgi Gerganovggerganov@gmail.com2025-09-29T16:49:11+03:00Georgi Gerganovggerganov@gmail.com2025-09-29T17:43:58+03:00sync : whisper.cpp (ggml/1359)
457c9b1c06467aca8d30fafcab51292bcb285df5b0db6ae75afb49b23db3dfbc2b01e8aabfb047e6880Daniel Beveniusdaniel.bevenius@gmail.com2025-09-26T17:34:42+02:00Georgi Gerganovggerganov@gmail.com2025-09-29T17:43:58+03:00ggml : remove -dev suffix from release version (ggml/1355)
458b6ae75afb49b23db3dfbc2b01e8aabfb047e6880b6dff20e2fe352093039e2359370c6bb2b505e0bDaniel Beveniusdaniel.bevenius@gmail.com2025-09-25T14:39:05+02:00Georgi Gerganovggerganov@gmail.com2025-09-29T17:43:58+03:00ggml : bump version to 0.9.3 (ggml/1353)
459b6dff20e2fe352093039e2359370c6bb2b505e0b2db78c75e4ef7497313fda6dbbb3fcaf9ca0752dGeorgi Gerganovggerganov@gmail.com2025-09-20T16:44:23+03:00Georgi Gerganovggerganov@gmail.com2025-09-29T17:43:58+03:00ggml : prepare for development of 0.9.2-dev
4602db78c75e4ef7497313fda6dbbb3fcaf9ca0752d02463ab27b1379ff5e3d936ce8b3bfd356872ea6Georgi Gerganovggerganov@gmail.com2025-09-20T16:44:23+03:00Georgi Gerganovggerganov@gmail.com2025-09-29T17:43:58+03:00ggml : bump version to 0.9.1
46102463ab27b1379ff5e3d936ce8b3bfd356872ea6adc76347d73b3d915e946efa5de8d0ad9f3904c2Rafal Lewczukrafal.lewczuk@gmail.com2025-09-29T13:17:09+02:00GitHubnoreply@github.com2025-09-29T13:17:09+02:00ggml-backend : add root cause in error message if loading backend library fails (#16172)
462adc76347d73b3d915e946efa5de8d0ad9f3904c23a2bdcda0b31e51db954551e0cd49424133a84f3Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-29T11:09:00+02:00GitHubnoreply@github.com2025-09-29T11:09:00+02:00ggml : check cuda and metal argsort limits and add test (#16323)
4633a2bdcda0b31e51db954551e0cd49424133a84f366bb7985c3fcefda7a74aae9f8321f70eb1646c4Aleksander Grygieraleksander.grygier@gmail.com2025-09-29T10:37:20+02:00GitHubnoreply@github.com2025-09-29T10:37:20+02:00Improve Mobile UI for dialogs and action dropdowns (#16222)
46466bb7985c3fcefda7a74aae9f8321f70eb1646c42f61c0f5bf8a620ca4c3872408803ab38cfb9613Pascaladmin@serveurperso.com2025-09-29T09:08:41+02:00GitHubnoreply@github.com2025-09-29T09:08:41+02:00fix: preserved zero values in chat settings inputs and textareas by switching to nullish coalescing for field values and default placeholders (#16312)
4652f61c0f5bf8a620ca4c3872408803ab38cfb96133ffd0fae473c954bb3e67526b31262048fb508d4Vinkalvinkal-chudgar@users.noreply.github.com2025-09-29T12:33:12+05:30GitHubnoreply@github.com2025-09-29T10:03:12+03:00llama-cli: prevent spurious assistant token (#16202)
4663ffd0fae473c954bb3e67526b31262048fb508d4a4a0aa5ea2a88e4858996199fefd5439f17b481cddh0chemist-mulches-39@icloud.com2025-09-29T01:30:45-05:00GitHubnoreply@github.com2025-09-29T09:30:45+03:00perplexity : show more kl-divergence data (#16321)
467a4a0aa5ea2a88e4858996199fefd5439f17b481c92cd103f627015a95f2107f1c27dc218c7b8ec64Georgi Gerganovggerganov@gmail.com2025-09-29T08:41:28+03:00GitHubnoreply@github.com2025-09-29T08:41:28+03:00ggml : fix dependencies for ggml_set_rows (#16318)
46892cd103f627015a95f2107f1c27dc218c7b8ec64b887d2f3413ac231e3cb5925260c39902af4a70cJeff Bolzjbolz@nvidia.com2025-09-28T23:50:37-05:00GitHubnoreply@github.com2025-09-29T06:50:37+02:00vulkan: Fix validation failure in quantized flash attention (#16292)
469b887d2f3413ac231e3cb5925260c39902af4a70cbd0af02fc96c2057726f33c0f0daf7bb8f3e462aSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-28T23:15:03+02:00GitHubnoreply@github.com2025-09-28T23:15:03+02:00ggml : fix GGML_F32_VEC_FMA argument order in ggml_vec_mad1_f32 (#16307)
470bd0af02fc96c2057726f33c0f0daf7bb8f3e462ad9e0e7c8194dfd7d23bf3a86608c9ece68d77c93crat0z11581854+crat0z@users.noreply.github.com2025-09-28T14:13:50-04:00GitHubnoreply@github.com2025-09-28T21:13:50+03:00common : fix reasoning before forced tool call via tool_choice = required (#16264)
471d9e0e7c8194dfd7d23bf3a86608c9ece68d77c930124ac989f7e7bf08803788f66dbe4106bdcdd58R0CKSTARyeahdongcn@gmail.com2025-09-28T22:38:15+08:00GitHubnoreply@github.com2025-09-28T16:38:15+02:00ci : fix musa docker build (#16306)
4720124ac989f7e7bf08803788f66dbe4106bdcdd582811c65286ae954bec87049f75b86dc022006dccAaron Teoaaron.teo1@ibm.com2025-09-28T19:25:58+08:00GitHubnoreply@github.com2025-09-28T19:25:58+08:00devops: switch to using ubuntu-22.04-s390x image (#16302)
4732811c65286ae954bec87049f75b86dc022006dccd8359f5fde480da030bf75c7711573c7c4d993baImad Saddik79410781+ImadSaddik@users.noreply.github.com2025-09-28T12:04:46+01:00GitHubnoreply@github.com2025-09-28T13:04:46+02:00Fixed a few typos in the README of the LLaMA.cpp HTTP Server [no ci] (#16297)
474d8359f5fde480da030bf75c7711573c7c4d993ba6a2c6145a0b91b40eb3c3dba7b20ccc4b270490fJeff Bolzjbolz@nvidia.com2025-09-28T01:38:37-05:00GitHubnoreply@github.com2025-09-28T08:38:37+02:00vulkan: 64-bit im2col (#16135)
4756a2c6145a0b91b40eb3c3dba7b20ccc4b270490f3b53634fe35771e2e318227aa81585726bae7234Georgi Gerganovggerganov@gmail.com2025-09-28T09:34:44+03:00GitHubnoreply@github.com2025-09-28T09:34:44+03:00metal : extend mat-mat multiplication support (#16225)
4763b53634fe35771e2e318227aa81585726bae72341384abf8b8d5894d32fada453ccf4d196ffba7deGeorgi Gerganovggerganov@gmail.com2025-09-28T09:34:05+03:00GitHubnoreply@github.com2025-09-28T09:34:05+03:00metal : fuse non-sequential nodes (#16102)
4771384abf8b8d5894d32fada453ccf4d196ffba7dee6d65fb02d553bd79cad94e517cdca18b687788dJeff Bolzjbolz@nvidia.com2025-09-27T20:36:34-05:00GitHubnoreply@github.com2025-09-27T20:36:34-05:00vulkan: handle mat_mul with A matrix > 4GB (#16176)
478e6d65fb02d553bd79cad94e517cdca18b687788d8656f5de688cddcaea1d6174535eb60ee23ef6a0Jeff Bolzjbolz@nvidia.com2025-09-27T16:43:39-04:00GitHubnoreply@github.com2025-09-27T22:43:39+02:00vulkan: support arbitrary KV dimension in flash attention (#16160)
4798656f5de688cddcaea1d6174535eb60ee23ef6a04807e8f96a61b2adccebd5e57444c94d18de7264Aclyaclysia@gmail.com2025-09-27T22:41:03+02:00GitHubnoreply@github.com2025-09-27T22:41:03+02:00vulkan : make the vulkan.hpp dynamic dispatcher instance private (#16224)
4804807e8f96a61b2adccebd5e57444c94d18de7264c0bfc57af421f8fd63c946c13b7666aed82560e2Aleksander Grygieraleksander.grygier@gmail.com2025-09-27T19:56:40+02:00GitHubnoreply@github.com2025-09-27T19:56:40+02:00Show message actions by default (#16289)
481c0bfc57af421f8fd63c946c13b7666aed82560e275a3a6c2cd0002ba40e2dcc92007bc9fdbc69f1aAman Guptaamangupta052@gmail.com2025-09-28T00:49:32+08:00GitHubnoreply@github.com2025-09-27T18:49:32+02:00CUDA: mul_mat_id for mmf for bs <= 64 for f16 and bs <= 32 for f32 (#16277)
48275a3a6c2cd0002ba40e2dcc92007bc9fdbc69f1a0499b29c6f64c705faaf5860dc4600fca23671f4Johannes Gäßlerjohannesg@5d6.de2025-09-27T18:45:07+02:00GitHubnoreply@github.com2025-09-27T18:45:07+02:00CUDA: refactor and deduplicate vector FA kernels (#16208)
4830499b29c6f64c705faaf5860dc4600fca23671f4234e2ff8ed09716fb553437596779399bee31b11Dmytro Minochkindmytro.minochkin@gmail.com2025-09-27T19:26:46+03:00GitHubnoreply@github.com2025-09-27T18:26:46+02:00vulkan: throw system error instead of SIGABRT during init on older devices (#16156)
484234e2ff8ed09716fb553437596779399bee31b113f81b4e91c1d5f098148af117e3f13cf4b077f52Adrien Gallouëtangt@huggingface.co2025-09-27T18:17:08+02:00GitHubnoreply@github.com2025-09-27T19:17:08+03:00server : remove old LLAMA_SERVER_SSL (#16290)
4853f81b4e91c1d5f098148af117e3f13cf4b077f52ace6a54565444b6377bee8e7ac693238e7766279Jeff Bolzjbolz@nvidia.com2025-09-27T06:36:11-04:00GitHubnoreply@github.com2025-09-27T12:36:11+02:00vulkan: support GET_ROWS for k-quants (#16235)
486ace6a54565444b6377bee8e7ac693238e776627972b24d96c6888c609d562779a23787304ae4609cAdrien Gallouëtangt@huggingface.co2025-09-27T11:12:46+02:00GitHubnoreply@github.com2025-09-27T12:12:46+03:00build : add LLAMA_OPENSSL option (#16287)
48772b24d96c6888c609d562779a23787304ae4609c624207e676ab5eb3ce7af631902bb45fb73a8359Vinkalvinkal-chudgar@users.noreply.github.com2025-09-27T02:58:29+05:30GitHubnoreply@github.com2025-09-26T23:28:29+02:00model : make minicpm embedding_scale, residual_scale and logit_scale optional with legacy defaults (#16273)
488624207e676ab5eb3ce7af631902bb45fb73a8359807e8c6d310952f2f5656afc63dab9d7083dcb5cAaron Teoaaron.teo1@ibm.com2025-09-27T02:03:33+08:00GitHubnoreply@github.com2025-09-27T02:03:33+08:00devops: add s390x & ppc64le CI (#15925)
489807e8c6d310952f2f5656afc63dab9d7083dcb5c1a189278944d030211f336e103e96b65f976c361Aleksander Grygieraleksander.grygier@gmail.com2025-09-26T19:25:29+02:00GitHubnoreply@github.com2025-09-26T19:25:29+02:00Enhance text file detection logic for file attachments (#16199)
4901a189278944d030211f336e103e96b65f976c361e0539eb6aed346d4b25a6ea019044e88771e7690Aleksander Grygieraleksander.grygier@gmail.com2025-09-26T18:35:42+02:00GitHubnoreply@github.com2025-09-26T18:35:42+02:00Allow viewing conversations even when llama server is down (#16255)
491e0539eb6aed346d4b25a6ea019044e88771e76905d0a40f390732cbf85d8a3b7b0fc3cbebffe780aIsaac McFadyenisaac@imcf.me2025-09-26T11:36:48-04:00GitHubnoreply@github.com2025-09-26T18:36:48+03:00webui: switch to hash-based routing (alternative of #16079) (#16157)
4925d0a40f390732cbf85d8a3b7b0fc3cbebffe780ad12a9836597b1ae4440d64d5a6ed727d27ac0702Aleksander Grygieraleksander.grygier@gmail.com2025-09-26T15:59:07+02:00GitHubnoreply@github.com2025-09-26T15:59:07+02:00Always show message actions for mobile UI + improvements for user message sizing (#16076)
493d12a9836597b1ae4440d64d5a6ed727d27ac0702cc1cfa277b3aae1f6cc9180472072336597a78d4Radoslav Gerganovrgerganov@gmail.com2025-09-26T16:09:34+03:00GitHubnoreply@github.com2025-09-26T16:09:34+03:00codeowners : add rgerganov as owner of RPC [no ci] (#16279)
494cc1cfa277b3aae1f6cc9180472072336597a78d454dbc37053f7d75ea5b0631d5ee88f19391e4314Aleksei Nikiforov103434461+AlekseiNikiforovIBM@users.noreply.github.com2025-09-26T15:00:44+02:00GitHubnoreply@github.com2025-09-26T15:00:44+02:00mtmd : fix uninitialized variable in bicubic_resize (#16275)
49554dbc37053f7d75ea5b0631d5ee88f19391e4314b995a10760cb93d23d617d76ecb82a5f95b5e0d3Georgi Gerganovggerganov@gmail.com2025-09-26T14:14:28+03:00GitHubnoreply@github.com2025-09-26T14:14:28+03:00metal : report OOM errors (#16274)
496b995a10760cb93d23d617d76ecb82a5f95b5e0d34710dd31bbcef79d04f85a3a6a8c7d9439c5c79aAdrien Gallouëtangt@huggingface.co2025-09-26T13:12:19+02:00GitHubnoreply@github.com2025-09-26T14:12:19+03:00common : use cpp-httplib as a cURL alternative for downloads (#16185)
4974710dd31bbcef79d04f85a3a6a8c7d9439c5c79a9b26511857ac09ae69ab485168fe2d3ee5fb1d6eAdrien Gallouëtangt@huggingface.co2025-09-26T12:39:35+02:00GitHubnoreply@github.com2025-09-26T13:39:35+03:00build : fix build-ios-device (#16257)
4989b26511857ac09ae69ab485168fe2d3ee5fb1d6e00217cd41388328c93e9e9644921c4319bb03bccAaron Teoaaron.teo1@ibm.com2025-09-26T18:27:25+08:00GitHubnoreply@github.com2025-09-26T13:27:25+03:00ggml-cpu: implement MXFP4 SIMD for s390x (#16193)
49900217cd41388328c93e9e9644921c4319bb03bcc3b337b01a1a85c2f5b49376e8c0bbdb3f521528eRadoslav Gerganovrgerganov@gmail.com2025-09-26T13:19:23+03:00GitHubnoreply@github.com2025-09-26T10:19:23Zci : create git tags for released docker images (#16008)
5003b337b01a1a85c2f5b49376e8c0bbdb3f521528ea86a580a6692edd1da076d5422f944c344b4fe5eDaniel Beveniusdaniel.bevenius@gmail.com2025-09-26T07:53:36+02:00GitHubnoreply@github.com2025-09-26T08:53:36+03:00codeowners : add danbev as owner of build-xcframework.sh [no ci] (#16268)
501a86a580a6692edd1da076d5422f944c344b4fe5e0f7c69689f4e681948a6005adea9bee9c08ff903R0CKSTARyeahdongcn@gmail.com2025-09-26T08:56:38+08:00GitHubnoreply@github.com2025-09-26T02:56:38+02:00musa: upgrade musa sdk to 4.3.0 (#16240)
5020f7c69689f4e681948a6005adea9bee9c08ff903835b2b915c52bcabcd688d025eacff9a07b65f52R0CKSTARyeahdongcn@gmail.com2025-09-26T08:56:10+08:00GitHubnoreply@github.com2025-09-26T02:56:10+02:00musa: fix build warnings (#15611)
503835b2b915c52bcabcd688d025eacff9a07b65f52b05a9d650f4da1e70e7e2cbe8fe44e61b39656dbSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-25T19:50:28+02:00GitHubnoreply@github.com2025-09-25T19:50:28+02:00model : add GroveMoE support (#15510)
504b05a9d650f4da1e70e7e2cbe8fe44e61b39656db27052978e487957b3d39e3bd8fc8ee4aa304bff9Aaron Teoaaron.teo1@ibm.com2025-09-25T23:38:10+08:00GitHubnoreply@github.com2025-09-25T23:38:10+08:00vendors: update miniaudio version (#16212)
50527052978e487957b3d39e3bd8fc8ee4aa304bff9077c94d0caf87fbd3cf3288dbb5c0fd9670294cfrtaluyevtaluyev@gmail.com2025-09-25T18:20:34+03:00GitHubnoreply@github.com2025-09-25T18:20:34+03:00readme : update bindings (#16144)
506077c94d0caf87fbd3cf3288dbb5c0fd9670294cfaa3ee0eb0b80efca126cedf9bcb4fb5864b46ce3Aman Guptaamangupta052@gmail.com2025-09-25T22:35:05+08:00GitHubnoreply@github.com2025-09-25T16:35:05+02:00CUDA: add a fused top-K MoE kernel (#16130)
507aa3ee0eb0b80efca126cedf9bcb4fb5864b46ce3d0991da39d3c39b3980c24cdfccb9a4c95a46870Daniel Beveniusdaniel.bevenius@gmail.com2025-09-25T12:02:36+02:00GitHubnoreply@github.com2025-09-25T12:02:36+02:00model-conversion : add embedding prompt file support (#15871)
508d0991da39d3c39b3980c24cdfccb9a4c95a46870aa719c2f886c445b78113bf1e5849f1fce4124a7Daniel Beveniusdaniel.bevenius@gmail.com2025-09-25T11:36:47+02:00GitHubnoreply@github.com2025-09-25T11:36:47+02:00server : add support for external server for tests (#16243)
509aa719c2f886c445b78113bf1e5849f1fce4124a74cdd0bb4537f9617e9efcfef6b9454fcefe2ff08junchao-zhao68935141+junchao-loongson@users.noreply.github.com2025-09-25T17:22:55+08:00GitHubnoreply@github.com2025-09-25T12:22:55+03:00ggml : fix loongarch lsx compilation error (#15864)
5104cdd0bb4537f9617e9efcfef6b9454fcefe2ff08b5bd037832bcb8ed3086dfe26ce9090bea989af1Johannes Gäßlerjohannesg@5d6.de2025-09-25T11:12:27+02:00GitHubnoreply@github.com2025-09-25T12:12:27+03:00docs: fix typo [no ci] (#16244)
511b5bd037832bcb8ed3086dfe26ce9090bea989af1dfcd53f7ecb9bb897a9d752d09c59d10be47237aDouglas Hanleythesecretaryofwar@gmail.com2025-09-25T03:53:09-05:00GitHubnoreply@github.com2025-09-25T11:53:09+03:00llama : add support for qwen3 reranker (#15824)
512dfcd53f7ecb9bb897a9d752d09c59d10be47237a4ea00794b8c995b6deaf4bac159c1778dc27419aGeorgi Gerganovggerganov@gmail.com2025-09-25T11:30:16+03:00GitHubnoreply@github.com2025-09-25T11:30:16+03:00metal : fuse NORM + MUL + ADD, support non-multiples of 4 (#16220)
5134ea00794b8c995b6deaf4bac159c1778dc27419a02a6a82ae7c7ddd1819ae26b0cc36675879aeceeGeorgi Gerganovggerganov@gmail.com2025-09-25T11:29:42+03:00GitHubnoreply@github.com2025-09-25T11:29:42+03:00metal : relax reorder conditions (#16216)
51402a6a82ae7c7ddd1819ae26b0cc36675879aeceec498fc82fe5b83fc8c6e1627286bdc1f93caddbfGeorgi Gerganovggerganov@gmail.com2025-09-25T11:29:08+03:00GitHubnoreply@github.com2025-09-25T11:29:08+03:00metal : restore im2col perf (#16219)
515c498fc82fe5b83fc8c6e1627286bdc1f93caddbfe7a5130a20cf45a3358308356c249734ca982143Radoslav Gerganovrgerganov@gmail.com2025-09-25T10:20:02+03:00GitHubnoreply@github.com2025-09-25T07:20:02Zrpc : use ggml logging facilities
516e7a5130a20cf45a3358308356c249734ca982143bee378e0988b44bbe93b0768208080951b312363Aaron Teoaaron.teo1@ibm.com2025-09-25T13:06:30+08:00GitHubnoreply@github.com2025-09-25T08:06:30+03:00codeowners: add ownership of zdnn backend [no ci] (#16232)
517bee378e0988b44bbe93b0768208080951b3123635fb557653b8756ac2b6c6a102b1e44abcadf552fEve139727413+netrunnereve@users.noreply.github.com2025-09-25T05:06:06ZGitHubnoreply@github.com2025-09-25T08:06:06+03:00ci: run the x64 and arm ci on the github machines instead (#16183)
5185fb557653b8756ac2b6c6a102b1e44abcadf552f4ae88d07d026e66b41e85afece74e88af54f4e66Aaron Teoaaron.teo1@ibm.com2025-09-25T11:36:30+08:00GitHubnoreply@github.com2025-09-25T11:36:30+08:00devops: fix s390x docker release failure (#16231)
5194ae88d07d026e66b41e85afece74e88af54f4e66e789095502b337690c7616db32d7c679a5bd2533Aaron Teoaaron.teo1@ibm.com2025-09-25T00:25:04+08:00GitHubnoreply@github.com2025-09-25T00:25:04+08:00codeowners: add ownership of zdnn backend [no ci] (#16229)
520e789095502b337690c7616db32d7c679a5bd2533f2a789e33490deb483a2694b066b37e45524bb79Johannes Gäßlerjohannesg@5d6.de2025-09-24T16:53:48+02:00GitHubnoreply@github.com2025-09-24T16:53:48+02:00llama: print memory breakdown on exit (#15860)
521f2a789e33490deb483a2694b066b37e45524bb793a599719673c850647e3bb911ed6d91109bb91d2Aclyaclysia@gmail.com2025-09-24T16:17:49+02:00GitHubnoreply@github.com2025-09-24T16:17:49+02:00ggml : split graph allocations according to backend max buffer size (#15815)
5223a599719673c850647e3bb911ed6d91109bb91d263b54c81a620981be020184ab99e63a8e50e47cbTarek Dakhrant.dakhran@gmail.com2025-09-24T13:42:26+02:00GitHubnoreply@github.com2025-09-24T13:42:26+02:00model : add label for LiquidAI LFM2-2.6B model (#16204)
523152729f8848b692e1ae53c197ccca92ef10a523bc0c59c1157b5aeded285a3b25d2463c866f583d3Uilian Riesuilianries@gmail.com2025-09-24T08:53:47+02:00GitHubnoreply@github.com2025-09-24T09:53:47+03:00common : add missing chrono header for common.cpp (#16211)
524c0c59c1157b5aeded285a3b25d2463c866f583d37735706b939847df2c6c00b44c36f95a20137c61Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-24T08:53:20+02:00GitHubnoreply@github.com2025-09-24T08:53:20+02:00codeowners : match all requirements files (#16214)
5257735706b939847df2c6c00b44c36f95a20137c614d9ea03d170cf504a40bf3a85788af84cccaee80Jie Fu (傅杰)jiefu@tencent.com2025-09-24T14:46:52+08:00GitHubnoreply@github.com2025-09-24T08:46:52+02:00model-conversion : run-org-model.py fails to run on mac m1 (#16213)
5264d9ea03d170cf504a40bf3a85788af84cccaee808ba548dae251f8f2e3834ed5226b76b6e4f4a485Daniel Beveniusdaniel.bevenius@gmail.com2025-09-24T08:10:09+02:00GitHubnoreply@github.com2025-09-24T08:10:09+02:00codeowners : use slash prefix for root files [no ci] (#16210)
5278ba548dae251f8f2e3834ed5226b76b6e4f4a485f505bd83ca7a43c4585ff3d59135e77eae9c793bJie Fu (傅杰)jiefu@tencent.com2025-09-24T12:19:23+08:00GitHubnoreply@github.com2025-09-24T06:19:23+02:00model-conversion : fix the make targets in the README.md (#16209)
528f505bd83ca7a43c4585ff3d59135e77eae9c793b0889589dbe8c67dd518c58a57afbb02dde2dccbeGeorgi Gerganovggerganov@gmail.com2025-09-23T20:41:40+03:00GitHubnoreply@github.com2025-09-23T20:41:40+03:00ci : disable AMD workflows + update NVIDIA workflows (#16200)
5290889589dbe8c67dd518c58a57afbb02dde2dccbe4e29084ba4104c4ea529fd3163bb6e76f64383dfGeorgi Gerganovggerganov@gmail.com2025-09-23T13:44:25+03:00GitHubnoreply@github.com2025-09-23T13:44:25+03:00ci : enable Vulkan workflow on Mac (#16194)
5304e29084ba4104c4ea529fd3163bb6e76f64383dff6b4af3d04763b1e0130f5b5fce19c4bc6f83f1cXiangyan Sunwishstudio@gmail.com2025-09-23T01:58:12-07:00GitHubnoreply@github.com2025-09-23T11:58:12+03:00ggml-cpu: Respect cpumask settings (#16164)
531f6b4af3d04763b1e0130f5b5fce19c4bc6f83f1c264f1b51872c125e23fa0ac1da5e2a1170de9a08Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-23T10:25:20+02:00GitHubnoreply@github.com2025-09-23T10:25:20+02:00ggml : fix uninitialized is_on_grid in quantize_row_iq3_xxs_impl (#15928)
532264f1b51872c125e23fa0ac1da5e2a1170de9a080bc7cc715472fe28822e57f036f0746592ed2c04Aaron Teoaaron.teo1@ibm.com2025-09-23T14:53:05+08:00GitHubnoreply@github.com2025-09-23T14:53:05+08:00zdnn: refactor codebase + add docs (#16178)
5330bc7cc715472fe28822e57f036f0746592ed2c044b9f4cb0f89a88de4bdf97727d0457b0c648804cDaniel Beveniusdaniel.bevenius@gmail.com2025-09-23T08:13:22+02:00GitHubnoreply@github.com2025-09-23T09:13:22+03:00codeowners : add @danbev to model-conversion example [no ci] (#16190)
5344b9f4cb0f89a88de4bdf97727d0457b0c648804c85e72271ba1ce78adf34fd8997803c991e617ca6Aaron Teoaaron.teo1@ibm.com2025-09-23T13:59:34+08:00GitHubnoreply@github.com2025-09-23T13:59:34+08:00devops: add s390x containers (#15915)
53585e72271ba1ce78adf34fd8997803c991e617ca61d0125bcf1cbd7195ad0faf826a20bc7cec7d3f4Daniel Beveniusdaniel.bevenius@gmail.com2025-09-23T05:59:03+02:00GitHubnoreply@github.com2025-09-23T05:59:03+02:00ggml-cpu : fix typo in gemm comments [no ci] (#16189)
5361d0125bcf1cbd7195ad0faf826a20bc7cec7d3f4351f3da39c85f59d581fc184f09283da7f099a3bGabe Goodhartghart@us.ibm.com2025-09-22T12:40:10-06:00GitHubnoreply@github.com2025-09-22T20:40:10+02:00feat: Add conversion support in GraniteHybrid for non-hybrid (all attn) (#16177)
537351f3da39c85f59d581fc184f09283da7f099a3b3ecb2f671a2f49d56357f99d135a94e841759178Haiyue Wanghaiyuewa@163.com2025-09-23T01:57:46+08:00GitHubnoreply@github.com2025-09-22T19:57:46+02:00clang-tidy : disable warning about performance enum size (#16127)
5383ecb2f671a2f49d56357f99d135a94e841759178432cf4304c5379bfbd1f3feed65ec205955b2f4bSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-22T19:13:00+02:00GitHubnoreply@github.com2025-09-22T19:13:00+02:00ggml : implement set_rows with i32 index (#16159)
539432cf4304c5379bfbd1f3feed65ec205955b2f4b37a23c17bdc9c99c9c6ad41168e4ced3724b72cdGeorgi Gerganovggerganov@gmail.com2025-09-22T18:20:21+03:00GitHubnoreply@github.com2025-09-22T18:20:21+03:00codeowners : update + cleanup (#16174)
54037a23c17bdc9c99c9c6ad41168e4ced3724b72cd138c87ce8bd558b2cc134ada7316a3dad8eb67acAdrien Gallouëtangt@huggingface.co2025-09-22T14:13:51+02:00GitHubnoreply@github.com2025-09-22T15:13:51+03:00common : enable `--offline` mode without curl support (#16137)
541138c87ce8bd558b2cc134ada7316a3dad8eb67acc6db9a10278f40b4b8d3f6931728f2cce2356daaQuentin Bramasquentin.bramas@gmail.com2025-09-22T10:53:13+02:00GitHubnoreply@github.com2025-09-22T11:53:13+03:00webui : fix handling incomplete chunks (#16107)
542c6db9a10278f40b4b8d3f6931728f2cce2356daad05affbab7d1364fa7ac06c03cd33f03235ae840GideonSerfgdserf.gs@gmail.com2025-09-22T10:49:58+02:00GitHubnoreply@github.com2025-09-22T11:49:58+03:00embedding : fix typos in README (#16171)
543d05affbab7d1364fa7ac06c03cd33f03235ae8404f324a556caa5be0be2261604ef4aab0faf36eb8Haiyue Wanghaiyuewa@163.com2025-09-22T16:48:42+08:00GitHubnoreply@github.com2025-09-22T11:48:42+03:00common : remove unused local variables (#16140)
5444f324a556caa5be0be2261604ef4aab0faf36eb8a71ae3ba7a857394103f12e30b387c48242c84f2Georgi Gerganovggerganov@gmail.com2025-09-22T11:12:37+03:00GitHubnoreply@github.com2025-09-22T11:12:37+03:00ggml : extend ggml_can_fuse to work with non-sequential nodes (#16123)
545a71ae3ba7a857394103f12e30b387c48242c84f205a2458121f9cdcc98ea6ffeeab6f23023136266Georgi Gerganovggerganov@gmail.com2025-09-22T11:12:09+03:00GitHubnoreply@github.com2025-09-22T11:12:09+03:00ggml : add ggml_op_is_empty (#16122)
54605a2458121f9cdcc98ea6ffeeab6f2302313626696fdca043be8fa733bbd59868a75517f92633376Xuan-Son Nguyenson@huggingface.co2025-09-22T15:10:58+07:00GitHubnoreply@github.com2025-09-22T11:10:58+03:00codeowners : update ownership for @ngxson and @allozuar (#16128)
54796fdca043be8fa733bbd59868a75517f92633376b2d980fce063fa3591c59ad50b946c8da66c294fShin-myoung-serprelent95@naver.com2025-09-22T17:04:01+09:00GitHubnoreply@github.com2025-09-22T10:04:01+02:00Vulkan: add conv_transpose_2d operation (#16022)
548b2d980fce063fa3591c59ad50b946c8da66c294f5c6106a696af2b00ce01d14ba525979d5a32b199Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-22T09:59:05+02:00GitHubnoreply@github.com2025-09-22T10:59:05+03:00codeowners : claim responsibility for ci, models, gguf-py and convert (#16124)
5495c6106a696af2b00ce01d14ba525979d5a32b199ec65fb52f0cc890e5085a6c5995ab08a156265fbGeorgi Gerganovggerganov@gmail.com2025-09-22T10:58:02+03:00GitHubnoreply@github.com2025-09-22T10:58:02+03:00contrib : update roles (#16113)
550ec65fb52f0cc890e5085a6c5995ab08a156265fb1d660d2fae42ea2e1d3569638e722bf7a37b6b19Georgi Gerganovggerganov@gmail.com2025-09-22T10:16:05+03:00GitHubnoreply@github.com2025-09-22T10:16:05+03:00ci : remove vulkaninfo calls (#16169)
5511d660d2fae42ea2e1d3569638e722bf7a37b6b19a20d810d79adfbf82e0eb703af6f27a0e2d1a539Georgi Gerganovggerganov@gmail.com2025-09-22T09:11:39+03:00GitHubnoreply@github.com2025-09-22T09:11:39+03:00ci : use smaller model (#16168)
552a20d810d79adfbf82e0eb703af6f27a0e2d1a5394d0a7cbc617e384fc355077a304c883b5c7d4fb6Jeff Bolzjbolz@nvidia.com2025-09-22T00:37:17-05:00GitHubnoreply@github.com2025-09-22T07:37:17+02:00vulkan: add RTE variants of exp shader (#16165)
5534d0a7cbc617e384fc355077a304c883b5c7d4fb69073a73d82a916cea0809de225ef5175c3a86e91Georgi Gerganovggerganov@gmail.com2025-09-22T08:31:40+03:00GitHubnoreply@github.com2025-09-22T08:31:40+03:00ci : adjust params for less runtime (#16167)
5549073a73d82a916cea0809de225ef5175c3a86e9151f5a45fbe575dcd54bdd2a339ef8e8424d1c12aRuben Ortlampicard12@live.de2025-09-22T07:22:43+02:00GitHubnoreply@github.com2025-09-22T07:22:43+02:00vulkan: vec dot matrix multiplication fix (#16151)
55551f5a45fbe575dcd54bdd2a339ef8e8424d1c12ac4510dc9374e17dcb8726902ab5216067a92b3d3lhezlih@qti.qualcomm.com2025-09-21T16:42:10-07:00GitHubnoreply@github.com2025-09-21T16:42:10-07:00opencl: fix concat crash on win arm64 with Adreno (#15944)
556c4510dc9374e17dcb8726902ab5216067a92b3d3da30ab5f8696cabb2d4620cdc0aa41a298c54fd6lhezlih@qti.qualcomm.com2025-09-21T14:48:44-07:00GitHubnoreply@github.com2025-09-21T14:48:44-07:00opencl: initial `q8_0` mv support (#15732)
557da30ab5f8696cabb2d4620cdc0aa41a298c54fd628baac9c9f491c872e2c37762d3bd90446b005e9Georgi Gerganovggerganov@gmail.com2025-09-21T19:00:27+03:00GitHubnoreply@github.com2025-09-21T19:00:27+03:00ci : add label for the RISC-V runner (#16150)
55828baac9c9f491c872e2c37762d3bd90446b005e91eeb523c3e0c7ffbd59469f5463dcbdecba3535eGeorgi Gerganovggerganov@gmail.com2025-09-21T16:50:45+03:00GitHubnoreply@github.com2025-09-21T16:50:45+03:00ci : migrate ggml ci to self-hosted runners (#16116)
5591eeb523c3e0c7ffbd59469f5463dcbdecba3535e5bb4a3edec297e74b0f7bd4ed5d0fdd12e28d858Giuseppe Scrivanogscrivan@redhat.com2025-09-21T08:31:55+02:00GitHubnoreply@github.com2025-09-21T08:31:55+02:00vulkan: optimize UMA buffer operations and fix driver hangs (#16059)
5605bb4a3edec297e74b0f7bd4ed5d0fdd12e28d8587f766929ca8e8e01dcceb1c526ee584f7e5e1408Jeff Bolzjbolz@nvidia.com2025-09-21T01:23:37-05:00GitHubnoreply@github.com2025-09-21T08:23:37+02:00vulkan: fix validation error about VK_PIPELINE_CREATE_CAPTURE_STATISTICS_BIT_KHR (#16086)
5617f766929ca8e8e01dcceb1c526ee584f7e5e1408405921dcefdd4e90ed948a4bf179007c2fa92b2dGeorgi Gerganovggerganov@gmail.com2025-09-20T12:55:47+03:00Georgi Gerganovggerganov@gmail.com2025-09-20T13:02:14+03:00sync : ggml
562405921dcefdd4e90ed948a4bf179007c2fa92b2dfa6383ca7e7ccb8ca3bdfeb37e348ddc4113aa26Daniel Beveniusdaniel.bevenius@gmail.com2025-09-16T06:16:52+02:00Georgi Gerganovggerganov@gmail.com2025-09-20T13:02:14+03:00ggml : introduce semantic versioning (ggml/1336)
563fa6383ca7e7ccb8ca3bdfeb37e348ddc4113aa26803dac2e48ef3ba26a504eb27c4e77ec2d21f7d0Gregor Jasnygjasny@googlemail.com2025-09-10T17:21:11+02:00Georgi Gerganovggerganov@gmail.com2025-09-20T13:02:14+03:00CUDA : conditionally add cuda architectures (ggml/1341)
564803dac2e48ef3ba26a504eb27c4e77ec2d21f7d0459c0c2c1a400f960d7b8e8d94d31a8426f80986Ruben Ortlampicard12@live.de2025-09-20T10:42:56+02:00GitHubnoreply@github.com2025-09-20T10:42:56+02:00vulkan: use vec dot for matrix matrix multiplications (#16056)
565459c0c2c1a400f960d7b8e8d94d31a8426f80986be79d9fdd95ab8955527c4aaa67b90e8b9516718Benni73313922+BenjaminBruenau@users.noreply.github.com2025-09-20T07:56:30+02:00GitHubnoreply@github.com2025-09-20T07:56:30+02:00server: fix SSE and OpenAI compatibility for error messages when streaming (#16109)
566be79d9fdd95ab8955527c4aaa67b90e8b9516718f432d8d83e7407073634c5e4fd81a3d23a10827fssweens1149151+ssweens@users.noreply.github.com2025-09-19T15:15:21-07:00GitHubnoreply@github.com2025-09-20T00:15:21+02:00llama-bench: add --devices and --list-devices support (#16039)
567f432d8d83e7407073634c5e4fd81a3d23a10827f4067f07fc5aa5722b87db303b561597004696f6cshun0958069181+shun095@users.noreply.github.com2025-09-20T00:57:30+09:00GitHubnoreply@github.com2025-09-19T09:57:30-06:00chat: Fix streaming parser for granite models (#15682)
5684067f07fc5aa5722b87db303b561597004696f6c4b8560ab56fdd9819358b47c338bbc8ec357c57eAleksander Grygieraleksander.grygier@gmail.com2025-09-19T09:52:27+02:00GitHubnoreply@github.com2025-09-19T09:52:27+02:00feat: Improve mobile UI for Settings Dialog (#16084)
5694b8560ab56fdd9819358b47c338bbc8ec357c57e0dd58b6877f3dc106593d6bc68d98305f553c566Xuan-Son Nguyenson@huggingface.co2025-09-19T13:02:51+07:00GitHubnoreply@github.com2025-09-19T13:02:51+07:00chat : fix build on arm64 (#16101)
5700dd58b6877f3dc106593d6bc68d98305f553c56669ffd891631befa9e6b485fd646a16dab4f2c007Xuan-Son Nguyenson@huggingface.co2025-09-19T11:31:56+07:00GitHubnoreply@github.com2025-09-19T06:31:56+02:00ggml : refactor forward_dup for cpu backend (#16062)
57169ffd891631befa9e6b485fd646a16dab4f2c007246c0d9c795fb25da8268625d597580155a3672fAdrien Gallouëtangt@huggingface.co2025-09-18T23:07:26+02:00GitHubnoreply@github.com2025-09-18T23:07:26+02:00ggml-amx : fix ggml_amx_init() on generic Linux (#16049)
572246c0d9c795fb25da8268625d597580155a3672f3edd87cd055a45d885fa914d879d36d33ecfc3e1Adrien Gallouëtangt@huggingface.co2025-09-18T23:07:18+02:00GitHubnoreply@github.com2025-09-18T23:07:18+02:00cmake : fix static linking for OpenMP on Unix-like systems (#16031)
5733edd87cd055a45d885fa914d879d36d33ecfc3e1c0b45097c33e2667a94444f08cc9e36bec0a5e2eShawn Gushawngu@qti.qualcomm.com2025-09-18T12:03:34-07:00GitHubnoreply@github.com2025-09-18T12:03:34-07:00opencl: optimize mxfp4 kernels (#16037)
574c0b45097c33e2667a94444f08cc9e36bec0a5e2e38dbdf4c057515ccea9bec0ca2518f86d5e4d28eJeff Bolzjbolz@nvidia.com2025-09-18T13:46:17-05:00GitHubnoreply@github.com2025-09-18T13:46:17-05:00rename optimize_graph to graph_optimize (#16082)
57538dbdf4c057515ccea9bec0ca2518f86d5e4d28e368560a1e3b9a3bc83af741b0b2bc9e46fb420d2Bowen Hanfancycode@gmail.com2025-09-18T11:26:03-07:00GitHubnoreply@github.com2025-09-18T20:26:03+02:00CUDA: Optimize PAD_REFLECT_1D (#15957)
576368560a1e3b9a3bc83af741b0b2bc9e46fb420d24ca088b036313c2d8e682f4cfeb7c29edd85d0b9Johannes Gäßlerjohannesg@5d6.de2025-09-18T19:28:32+02:00GitHubnoreply@github.com2025-09-18T19:28:32+02:00CUDA: fix compilation on CC 6.0 (#16091)
5774ca088b036313c2d8e682f4cfeb7c29edd85d0b9703f9e32c4eb3166f8d63007c26e31a1466c21afEric Curtineric.curtin@docker.com2025-09-18T16:22:50+01:00GitHubnoreply@github.com2025-09-18T16:22:50+01:00Add resumable downloads for llama-server model loading (#15963)
578703f9e32c4eb3166f8d63007c26e31a1466c21afad6bd9083bc346afd4ecd2fbb83e0306c0141c99Georgi Gerganovggerganov@gmail.com2025-09-18T16:28:41+03:00GitHubnoreply@github.com2025-09-18T16:28:41+03:00metal : use function constants for mul_mv_ext kernels (#16074)
579ad6bd9083bc346afd4ecd2fbb83e0306c0141c992b6b55a59f590a1e1eb2dcd09d5b8b6feb9cc748Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-18T13:28:22+02:00GitHubnoreply@github.com2025-09-18T13:28:22+02:00cuda : add missing F32<->I32 entries in ggml_cuda_cpy_fn (#16060)
5802b6b55a59f590a1e1eb2dcd09d5b8b6feb9cc748e58174cecbc45bf79bf653cd2c984395940c6ef4Radoslav Gerganovrgerganov@gmail.com2025-09-18T13:36:57+03:00GitHubnoreply@github.com2025-09-18T10:36:57Zserver : include usage statistics only when user request them (#16052)
581e58174cecbc45bf79bf653cd2c984395940c6ef4b213fce89bee8cb56b587b91e15a4278f8ed0180Georgi Gerganovggerganov@gmail.com2025-09-18T12:47:56+03:00GitHubnoreply@github.com2025-09-18T12:47:56+03:00llama : bump max seq limit from 64 to 256 (#15916)
582b213fce89bee8cb56b587b91e15a4278f8ed0180e00f3fd8fff2cf5a8c8c9f475034bd089c8bcce4Georgi Gerganovggerganov@gmail.com2025-09-18T12:33:45+03:00GitHubnoreply@github.com2025-09-18T12:33:45+03:00metal : improve F32, F16 and BF16 mat-vec multiplication (#16057)
583e00f3fd8fff2cf5a8c8c9f475034bd089c8bcce4f2f28380ea68939c0481df3e4216b698824a59dfJhen-Jie Hongiainst0409@gmail.com2025-09-18T15:06:48+08:00GitHubnoreply@github.com2025-09-18T10:06:48+03:00metal : avoid call free for non-owned buffer (#16067)
584f2f28380ea68939c0481df3e4216b698824a59df62c3b645c56ad5288aa401901f043b050edac5a2Georgi Gerganovggerganov@gmail.com2025-09-18T10:03:24+03:00GitHubnoreply@github.com2025-09-18T10:03:24+03:00metal : handle nil cv during pipeline creation (#16065)
58562c3b645c56ad5288aa401901f043b050edac5a2d304f459d8619399eb232b1e3689379306f439d3Chenguang Li757486878@qq.com2025-09-18T09:26:33+08:00GitHubnoreply@github.com2025-09-18T09:26:33+08:00CANN: Remove print (#16044)
586d304f459d8619399eb232b1e3689379306f439d30320ac5264279d74f8ee91bafa6c90e9ab9bbb91Reese Levinereeselevine1@gmail.com2025-09-17T13:09:40-07:00GitHubnoreply@github.com2025-09-17T13:09:40-07:00GGML WebGPU: Support for ADD, MUL, RMS_NORM, GET_ROWS operators (#16018)
5870320ac5264279d74f8ee91bafa6c90e9ab9bbb91a7a98e0fffed794396b3fbad4dcdbbc184963645Georgi Gerganovggerganov@gmail.com2025-09-17T20:38:12+03:00GitHubnoreply@github.com2025-09-17T20:38:12+03:00metal : refactor + optimize v2 (#15995)
588a7a98e0fffed794396b3fbad4dcdbbc1849636458f8f2274ee3601fecf6e2d57b52f701c81bede21Aleksander Grygieraleksander.grygier@gmail.com2025-09-17T19:29:13+02:00GitHubnoreply@github.com2025-09-17T19:29:13+02:00SvelteKit-based WebUI (#14839)
5898f8f2274ee3601fecf6e2d57b52f701c81bede21c959b676be29e93f8dbc3bd6056ceba812a9eb72Xuan-Son Nguyenson@huggingface.co2025-09-18T00:18:21+07:00GitHubnoreply@github.com2025-09-17T19:18:21+02:00convert : add Llama4ForCausalLM (#16042)
590c959b676be29e93f8dbc3bd6056ceba812a9eb72cd08fc3ecc0264b4414b68af3874a6c689ed60c1Johannes Gäßlerjohannesg@5d6.de2025-09-17T15:32:42+02:00GitHubnoreply@github.com2025-09-17T15:32:42+02:00CUDA: fix FA occupancy, optimize tile kernel (#15982)
591cd08fc3ecc0264b4414b68af3874a6c689ed60c1cb5bb6cc05119c24e7711ca2956cd0e6d409d396David Ribeiro Alvesdavidralves@gmail.com2025-09-17T01:08:02-07:00GitHubnoreply@github.com2025-09-17T11:08:02+03:00common : Fix corrupted memory error on json grammar initialization (#16038)
592cb5bb6cc05119c24e7711ca2956cd0e6d409d396a91d035b901e8a9edf810f63d130ee49adf27be2Eve139727413+netrunnereve@users.noreply.github.com2025-09-17T07:35:37ZGitHubnoreply@github.com2025-09-17T09:35:37+02:00vulkan: automatically remove unsupported devices (#15976)
593a91d035b901e8a9edf810f63d130ee49adf27be2745cbcf2fe1eb88f8db615ac622f0b944d924ad6Daniel Beveniusdaniel.bevenius@gmail.com2025-09-17T09:34:09+02:00GitHubnoreply@github.com2025-09-17T09:34:09+02:00ci : revert back to macos-13 for macOS-latest-cmake-x64 (#16040)
594745cbcf2fe1eb88f8db615ac622f0b944d924ad61cbd80f8cf80a817715b1ccc5680fe2a3c5172c8Jie Fu (傅杰)jiefu@tencent.com2025-09-17T15:30:55+08:00GitHubnoreply@github.com2025-09-17T09:30:55+02:00llama-quant : fix the verification of attention layers for encoder-decoder models (#16023)
5951cbd80f8cf80a817715b1ccc5680fe2a3c5172c885286f354813056f6c835046c0acfa3bf6ba9432Jie Fu (傅杰)jiefu@tencent.com2025-09-17T15:29:00+08:00GitHubnoreply@github.com2025-09-17T10:29:00+03:00examples : support encoder-decoder models in the simple example (#16002)
59685286f354813056f6c835046c0acfa3bf6ba9432d5fabe3682de515fd09d6c981f7a0d1b75614455Shane Ashanea@allenai.org2025-09-17T00:01:58-07:00GitHubnoreply@github.com2025-09-17T09:01:58+02:00model : add OLMo3 support (#16015)
597d5fabe3682de515fd09d6c981f7a0d1b756144558ff206097c2bf3ca1c7aa95f9d6db779fc7bdd68Chenguang Li757486878@qq.com2025-09-17T14:33:08+08:00GitHubnoreply@github.com2025-09-17T14:33:08+08:00CANN: Optimize ggml_cann_set_device (#15935)
5988ff206097c2bf3ca1c7aa95f9d6db779fc7bdd6877475530b8bbea3bf578632507e1284cdfe2c8c0jacekpoplawski67507230+jacekpoplawski@users.noreply.github.com2025-09-16T16:17:08+02:00GitHubnoreply@github.com2025-09-16T16:17:08+02:00llama-bench: add --n-cpu-moe support (#15952)
59977475530b8bbea3bf578632507e1284cdfe2c8c03913f8730ec6d6245480affc30ae3049107956f4Daniel Beveniusdaniel.bevenius@gmail.com2025-09-16T15:27:52+02:00GitHubnoreply@github.com2025-09-16T15:27:52+02:00ci : use macos-latest for arm64 webgpu build (#16029)
6003913f8730ec6d6245480affc30ae3049107956f476888d202ed2b835ae19ea9f9db6baf39e419297Daniel Beveniusdaniel.bevenius@gmail.com2025-09-16T15:25:57+02:00GitHubnoreply@github.com2025-09-16T15:25:57+02:00ggml : fix padding in timestep embedding kernels (#15932)
60176888d202ed2b835ae19ea9f9db6baf39e419297f1fbffb5c0b34b2a68febb7da3fd0f8333f1ed4cDaniel Beveniusdaniel.bevenius@gmail.com2025-09-16T13:41:38+02:00GitHubnoreply@github.com2025-09-16T13:41:38+02:00ci : upload xcframework artifact from ios-xcode-build job (#16010)
602f1fbffb5c0b34b2a68febb7da3fd0f8333f1ed4c51abc96bdc52ba8cd6ad78dcf12ed9a041d7b442Bowen Hanfancycode@gmail.com2025-09-15T23:59:19-07:00GitHubnoreply@github.com2025-09-16T08:59:19+02:00fix: apply clang-format to CUDA macros (#16017)
60351abc96bdc52ba8cd6ad78dcf12ed9a041d7b44207808ebb07e2b1aa19032705e332679ddf967614Daniel Beveniusdaniel.bevenius@gmail.com2025-09-16T05:57:16+02:00GitHubnoreply@github.com2025-09-16T05:57:16+02:00ci : update macos-latest* jobs to use macos-latest (#15938)
60407808ebb07e2b1aa19032705e332679ddf9676146d758839ff741d4966ca92b7f801b7a8b5b96364Yuri Khrustalevykhrustalev@users.noreply.github.com2025-09-15T22:54:44-04:00GitHubnoreply@github.com2025-09-16T09:54:44+07:00cmake : Do not install tools on iOS targets (#15903)
6056d758839ff741d4966ca92b7f801b7a8b5b963643d4053f77f0f78ee2b791088c02af653ebee42ddAman Guptaamangupta052@gmail.com2025-09-16T10:38:28+08:00GitHubnoreply@github.com2025-09-16T10:38:28+08:00Add LLaDA-7b-MoE diffusion model (#16003)
6063d4053f77f0f78ee2b791088c02af653ebee42dddc381aa9a6dc45f00673471d34b8bddd30e77570Jake Karnesjake.karnes@gmail.com2025-09-15T16:28:31-06:00GitHubnoreply@github.com2025-09-16T00:28:31+02:00CUDA: fix im2col_3d to respect non-contiguous inputs (views) (#15956)
607dc381aa9a6dc45f00673471d34b8bddd30e7757010d197409bd9537ff302ad09966fe406882fef9dDiego Devesaslarengh@gmail.com2025-09-15T14:38:52-07:00GitHubnoreply@github.com2025-09-15T23:38:52+02:00docker : enable rocWMMA in ROCm images, add gfx1151 (#15997)
60810d197409bd9537ff302ad09966fe406882fef9db907255f4bd169b0dc7dca9553b4c54af5170865Diego Devesaslarengh@gmail.com2025-09-15T14:38:42-07:00GitHubnoreply@github.com2025-09-15T23:38:42+02:00releases : switch to rocWMMA develop branch, add gfx1151 (#15992)
609b907255f4bd169b0dc7dca9553b4c54af517086528c39da7c645185ade5436767929d7ec33006033yael-works106673277+yael-works@users.noreply.github.com2025-09-15T19:51:35+03:00GitHubnoreply@github.com2025-09-15T18:51:35+02:00SYCL: Add COUNT_EQUAL operator support (#15991)
61028c39da7c645185ade5436767929d7ec33006033106220562aca42b6738b8f51acfce0db1b8a2fb6Nikolay Popov131475237+npopov-vst@users.noreply.github.com2025-09-15T13:08:30+03:00GitHubnoreply@github.com2025-09-15T11:08:30+01:00llama-run: Fix model download on Windows (#15988)
611106220562aca42b6738b8f51acfce0db1b8a2fb6a68f31edd71cc39141113f05f7133a3e9ece8c61Aman Guptaamangupta052@gmail.com2025-09-15T17:35:11+08:00GitHubnoreply@github.com2025-09-15T17:35:11+08:00CUDA: some micro-optimizations in mmf.cuh for mul_mat_id (#15926)
612a68f31edd71cc39141113f05f7133a3e9ece8c61b8e09f08b9a91c0401bc67d17a17c90756420346ddh0chemist-mulches-39@icloud.com2025-09-15T02:54:57-05:00GitHubnoreply@github.com2025-09-15T09:54:57+02:00fix KLD percentile output (#15999)
613b8e09f08b9a91c0401bc67d17a17c907564203466c019cb04e86e2dacfe62ce7666c64e9717dde1fSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-14T23:00:59+02:00GitHubnoreply@github.com2025-09-14T23:00:59+02:00model : add grok-2 support (#15539)
6146c019cb04e86e2dacfe62ce7666c64e9717dde1f9dcd200d57bc6f05a59a6a8df361d5d183af4124Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-14T21:17:04+02:00GitHubnoreply@github.com2025-09-14T21:17:04+02:00server : only attempt to enable thinking if using jinja (#15967)
6159dcd200d57bc6f05a59a6a8df361d5d183af41240fa154e3502e940df914f03b41475a2b80b985b0Georgi Gerganovggerganov@gmail.com2025-09-14T22:02:32+03:00GitHubnoreply@github.com2025-09-14T22:02:32+03:00metal : remove memory pools (#15966)
6160fa154e3502e940df914f03b41475a2b80b985b0261e6a20ffdb79c4875e674b4f6b514bc73cff8fAdamchanneladam@users.noreply.github.com2025-09-15T04:43:54+10:00GitHubnoreply@github.com2025-09-14T20:43:54+02:00rocm.Dockerfile: added gfx1200,gfx1201 architectures to support AMD Radeon RX 9000 series (#15994)
617261e6a20ffdb79c4875e674b4f6b514bc73cff8fa0e13dcbe5bae7025660349ef3e4ead060e507f2Ruben Ortlampicard12@live.de2025-09-14T16:56:28+02:00GitHubnoreply@github.com2025-09-14T16:56:28+02:00Vulkan: Clean up mul_mm shader (#15987)
618a0e13dcbe5bae7025660349ef3e4ead060e507f2a14bd350141fb42b8bf2dd2342cebc27bfdce399lcylcy0321@users.noreply.github.com2025-09-14T22:20:35+08:00GitHubnoreply@github.com2025-09-14T07:20:35-07:00build: fix the build failures of Windows HIP release job (#15984)
619a14bd350141fb42b8bf2dd2342cebc27bfdce399918b26f197f55d5d562446dfc876d0e637929d07Georgi Gerganovggerganov@gmail.com2025-09-14T15:33:22+03:00GitHubnoreply@github.com2025-09-14T15:33:22+03:00metal : fix kernel requirements (#15983)
620918b26f197f55d5d562446dfc876d0e637929d079ecb88434644c865232bb665d2f6f05049fc6456Radoslav Gerganovrgerganov@gmail.com2025-09-14T12:28:18+03:00GitHubnoreply@github.com2025-09-14T12:28:18+03:00rpc : fix regression when --device is used (#15981)
6219ecb88434644c865232bb665d2f6f05049fc6456d1c6f11f47bae063a68a6be8e4830c060a11b7bdDiego Devesaslarengh@gmail.com2025-09-14T02:21:59-07:00GitHubnoreply@github.com2025-09-14T02:21:59-07:00releases : update ROCM, add gfx1200, gfx1201, gfx1151 (#15972)
622d1c6f11f47bae063a68a6be8e4830c060a11b7bd6380d6a3e709cb02a8695afdd96b40e674477332Radoslav Gerganovrgerganov@gmail.com2025-09-14T12:10:07+03:00GitHubnoreply@github.com2025-09-14T12:10:07+03:00doc : update documentation for --tensor-split (#15980)
6236380d6a3e709cb02a8695afdd96b40e674477332aa0c461efe3603639af1a1defed2438d9c16ca0fAaron Teoaaron.teo1@ibm.com2025-09-14T13:37:03+08:00GitHubnoreply@github.com2025-09-14T13:37:03+08:00ggml-zdnn: rm user mapped buffers (#15965)
624aa0c461efe3603639af1a1defed2438d9c16ca0fb9c9c9f789cd57fb0b28b25223305613cd90fa10Jeff Bolzjbolz@nvidia.com2025-09-13T16:29:43+01:00GitHubnoreply@github.com2025-09-13T17:29:43+02:00vulkan: fix failing dequant shaders (#15862)
625b9c9c9f789cd57fb0b28b25223305613cd90fa1050f4281a6f5c3a5d68bdeb12f904fa01e0e2ba91Jeff Bolzjbolz@nvidia.com2025-09-13T16:23:30+01:00GitHubnoreply@github.com2025-09-13T17:23:30+02:00vulkan: initialize vulkan-hpp to allow using extension function pointers (#15705)
62650f4281a6f5c3a5d68bdeb12f904fa01e0e2ba9155758b00cae1451a0d789c50a5eb8c9bee42b9a0Diego Devesaslarengh@gmail.com2025-09-13T07:49:49-07:00GitHubnoreply@github.com2025-09-13T16:49:49+02:00llama : allow using iGPUs with --device (#15951)
62755758b00cae1451a0d789c50a5eb8c9bee42b9a0f161463a54d9f93d41246286aa4a9569a91d804dGeorgi Gerganovggerganov@gmail.com2025-09-13T16:24:22+03:00GitHubnoreply@github.com2025-09-13T16:24:22+03:00metal : refactor kernel loading (#15964)
628f161463a54d9f93d41246286aa4a9569a91d804d84d7b2fca11d1be118ce776f6d72a486c4883b74Georgi Gerganovggerganov@gmail.com2025-09-13T13:54:28+03:00GitHubnoreply@github.com2025-09-13T13:54:28+03:00metal : allow ops to run concurrently (#15929)
62984d7b2fca11d1be118ce776f6d72a486c4883b7440be51152d4dc2d47444a4ed378285139859895bGeorgi Gerganovggerganov@gmail.com2025-09-13T12:45:04+03:00GitHubnoreply@github.com2025-09-13T12:45:04+03:00metal : fix memory leaks (#15962)
63040be51152d4dc2d47444a4ed378285139859895b4bf5549269d99bac936a65892da2312cc71b2421Aaron Teoaaron.teo1@ibm.com2025-09-13T02:39:52+08:00GitHubnoreply@github.com2025-09-13T02:39:52+08:00ggml-zdnn: fix #15414, activate FP16 and BF16 acceleration and incorrect zTensor free (#15839)
6314bf5549269d99bac936a65892da2312cc71b2421f4e664f838cc65d89fa845c48c372e43852112e4Eric Curtineric.curtin@docker.com2025-09-12T16:31:50+01:00GitHubnoreply@github.com2025-09-12T16:31:50+01:00Add docker protocol support for llama-server model loading (#15790)
632f4e664f838cc65d89fa845c48c372e43852112e4f088b6a84f6aed0abb619dbb9d375e726fc888a6Haiyue Wanghaiyuewa@163.com2025-09-12T23:16:32+08:00GitHubnoreply@github.com2025-09-12T18:16:32+03:00context : remove redundant explicit casting to the same type (#15948)
633f088b6a84f6aed0abb619dbb9d375e726fc888a6304ac5693d1e2124f83e0584bc5eea6311d3d3b4Georgi Gerganovggerganov@gmail.com2025-09-12T17:02:55+03:00GitHubnoreply@github.com2025-09-12T17:02:55+03:00server : adjust prompt similarity thold + add logs (#15913)
634304ac5693d1e2124f83e0584bc5eea6311d3d3b46c88ad8fa741a2182a2dcf8c5353ae4b5c66e656Ruben Ortlampicard12@live.de2025-09-12T13:24:21+02:00GitHubnoreply@github.com2025-09-12T13:24:21+02:00Vulkan iGPU device selection overhaul and PCI ID API support (#15947)
6356c88ad8fa741a2182a2dcf8c5353ae4b5c66e656704d90c987cdf00751567b2088c4e54742aa2d3fMathieu Baudiermbaudier@argeo.org2025-09-12T09:06:20+02:00GitHubnoreply@github.com2025-09-12T09:06:20+02:00vulkan: Make device memory check more portable (#15939)
636360d6533db39e11577afe9b0aece20c6b5ddaf1f0e6ff0046f4a2983b2c77950aa75960fe4b4f0e2Diego Devesaslarengh@gmail.com2025-09-11T13:47:38-07:00GitHubnoreply@github.com2025-09-11T22:47:38+02:00ggml-backend : add GGML_BACKEND_DEVICE_TYPE_IGPU device type (#15797)
6370e6ff0046f4a2983b2c77950aa75960fe4b4f0e2df082f56309073ecf885eceaa21b86e8a487e61bJohannes Gäßlerjohannesg@5d6.de2025-09-11T21:19:58+02:00GitHubnoreply@github.com2025-09-11T21:19:58+02:00CUDA: larger SRAM reads for tile FA, AMD FP16 dot (#15927)
638df082f56309073ecf885eceaa21b86e8a487e61b24a6734daf6932ff29ba8c1ff0245c51d76f783eddh0chemist-mulches-39@icloud.com2025-09-11T12:12:34-05:00GitHubnoreply@github.com2025-09-11T19:12:34+02:00nitpick : correct MB to MiB (#15934)
63924a6734daf6932ff29ba8c1ff0245c51d76f783e2b3efea9a4d91216850856fbb77075db26f6a6ebDaniel Beveniusdaniel.bevenius@gmail.com2025-09-11T15:39:12+02:00GitHubnoreply@github.com2025-09-11T14:39:12+01:00ggml-cpu : add check for ARM MATMUL_INT8/i8mm support (#15922)
6402b3efea9a4d91216850856fbb77075db26f6a6ebc0389dba43d50695f9d3f57dd1f1a14cbefc100cCharles Xucharles.xu@arm.com2025-09-11T12:45:40+02:00GitHubnoreply@github.com2025-09-11T12:45:40+02:00kleidiai: fix GGML_ASSERT(*cur_backend_id != -1) failed (#15614)
641c0389dba43d50695f9d3f57dd1f1a14cbefc100c00681dfc16ba4cebb9c7fbd2cf2656e06a0692a4hipuddinghuafengchun@gmail.com2025-09-11T15:59:37+08:00GitHubnoreply@github.com2025-09-11T15:59:37+08:00CANN: Disable acl_graph for prefill stage (#15933)
64200681dfc16ba4cebb9c7fbd2cf2656e06a0692a44f658855fa8f2e42b7ed9a5b298fa39a2e39b096Oliver Simonsosimons@nvidia.com2025-09-10T22:04:03+02:00GitHubnoreply@github.com2025-09-10T22:04:03+02:00CUDA: Add `fastdiv` to `k_bin_bcast*`, giving 1-3% E2E performance (#15872)
6434f658855fa8f2e42b7ed9a5b298fa39a2e39b0966ab397e12ba8e9f776341cdae68f7ffb2f8d2cdeJie Fu (傅杰)jiefu@tencent.com2025-09-11T02:51:51+08:00GitHubnoreply@github.com2025-09-10T20:51:51+02:00llama : support T5 models with unequal number of encoder-decoder layers (#15909)
6446ab397e12ba8e9f776341cdae68f7ffb2f8d2cde9de447d94e1ae9d1a36e5a2e5bf47483352c0d9cSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-10T19:08:59+02:00GitHubnoreply@github.com2025-09-10T19:08:59+02:00graph : support non-contiguous Q in build_attn_mha (#15908)
6459de447d94e1ae9d1a36e5a2e5bf47483352c0d9c0f0a3c2851134d49955f3c85afbb0b1bb47c3e07Daniel Beveniusdaniel.bevenius@gmail.com2025-09-10T17:31:40+02:00GitHubnoreply@github.com2025-09-10T17:31:40+02:00ggml-cpu : fix padding in ggml_timestep_embedding (#15917)
6460f0a3c2851134d49955f3c85afbb0b1bb47c3e0733daece86b65607451d0d4378d2d04ba6a20ad55Georgi Gerganovggerganov@gmail.com2025-09-10T17:52:35+03:00GitHubnoreply@github.com2025-09-10T17:52:35+03:00metal : make the backend async (#15906)
64733daece86b65607451d0d4378d2d04ba6a20ad55e7b6d83b524bbc24a1343d862de6dba8e8eddbd6Daniel Beveniusdaniel.bevenius@gmail.com2025-09-10T15:39:57+02:00GitHubnoreply@github.com2025-09-10T15:39:57+02:00ci : add caching for ROCm installation in release workflow (#15924)
648e7b6d83b524bbc24a1343d862de6dba8e8eddbd62cfef4d117d67ab1dec002915b48a15d11ee1973Daniel Beveniusdaniel.bevenius@gmail.com2025-09-10T14:17:09+02:00GitHubnoreply@github.com2025-09-10T14:17:09+02:00tests : filter out no-ops from coverage report (#15900)
6492cfef4d117d67ab1dec002915b48a15d11ee197309e72a037c77b6823fde9e1e4cf28e815b9c41abj-kdev@j-k.io2025-09-10T12:51:28+01:00GitHubnoreply@github.com2025-09-10T14:51:28+03:00media : add transparent icon svg and png [no ci] (#15891)
65009e72a037c77b6823fde9e1e4cf28e815b9c41ab10d8b2b6b0ac2cae252a80b4daea5da55ab63c2fJessejesse@createthis.com2025-09-10T07:28:47-04:00GitHubnoreply@github.com2025-09-10T14:28:47+03:00gitignore : Ignore vim swap files in tests (#15901)
65110d8b2b6b0ac2cae252a80b4daea5da55ab63c2f28b5f190ef1dbea5edf82dbc8b4407b721fadd13Chenguang Li757486878@qq.com2025-09-10T18:42:00+08:00GitHubnoreply@github.com2025-09-10T18:42:00+08:00CANN: Add ROPE sin/cos cache for reuse (#15912)
65228b5f190ef1dbea5edf82dbc8b4407b721fadd1386587da03bd78df8f4e7d8b111a0c1d2494d6ed0Chenguang Li757486878@qq.com2025-09-10T15:29:12+08:00GitHubnoreply@github.com2025-09-10T15:29:12+08:00CANN: implement LRU cache for ACL graphs (#15814)
65386587da03bd78df8f4e7d8b111a0c1d2494d6ed0ff02caf9eed261423289d1531a56536fbf57bfc2Daniel Beveniusdaniel.bevenius@gmail.com2025-09-10T05:33:58+02:00GitHubnoreply@github.com2025-09-10T05:33:58+02:00llama : check returned fn ptrs from ggml_backend_reg_get_proc_address (#15893)
654ff02caf9eed261423289d1531a56536fbf57bfc2ae355f6f7108540297d8b7f7ae71d20fe610a0b7Daniel Beveniusdaniel.bevenius@gmail.com2025-09-10T05:23:19+02:00GitHubnoreply@github.com2025-09-10T05:23:19+02:00ci : cache ROCm installation in windows-latest-cmake-hip (#15887)
655ae355f6f7108540297d8b7f7ae71d20fe610a0b74f63cd705c7b6f457f36c63fdc053e07f6f3cc6bRuben Ortlampicard12@live.de2025-09-09T22:26:03+02:00GitHubnoreply@github.com2025-09-09T22:26:03+02:00vulkan: throw the oom error instead of no memory type found (#15905)
6564f63cd705c7b6f457f36c63fdc053e07f6f3cc6b17bc5a815f0bebc1844c99796760cd7df5e9b8d9Jeff Bolzjbolz@nvidia.com2025-09-09T07:41:15-05:00GitHubnoreply@github.com2025-09-09T14:41:15+02:00vulkan: Fix OOB accesses in soft_max_back (#15861)
65717bc5a815f0bebc1844c99796760cd7df5e9b8d9ed54e32558ffe0f2c3b1d31f3227211fae05bd49Johannes Gäßlerjohannesg@5d6.de2025-09-09T14:04:43+02:00GitHubnoreply@github.com2025-09-09T14:04:43+02:00HIP: use v_dot2_f32_f16 instruction for FA (#15884)
658ed54e32558ffe0f2c3b1d31f3227211fae05bd49a972faebed5fdc4a3d2a844d92d476058c02e02dlksj92hs134250687+lksj92hs@users.noreply.github.com2025-09-09T15:01:15+03:00GitHubnoreply@github.com2025-09-09T14:01:15+02:00Workaround for subgroup arithmetic failing on MoltenVK with AMD GPUs (issue 15846) (#15886)
659a972faebed5fdc4a3d2a844d92d476058c02e02d550cf726e133fd0a069d991287fd3a2a3e3e1cbdAman Guptaamangupta052@gmail.com2025-09-09T14:38:02+08:00GitHubnoreply@github.com2025-09-09T14:38:02+08:00CUDA: Add mul_mat_id support for the mmf kernel (#15767)
660550cf726e133fd0a069d991287fd3a2a3e3e1cbdc252ce67c4b99f056eeb18d42a289e28fee03475Johannes Gäßlerjohannesg@5d6.de2025-09-09T08:11:01+02:00GitHubnoreply@github.com2025-09-09T08:11:01+02:00CUDA: fix GET_ROWS for large tensors (#15882)
661c252ce67c4b99f056eeb18d42a289e28fee0347570cd37dbbebdb7a2f84f08207f18eabb0b291a55Georgi Gerganovggerganov@gmail.com2025-09-09T08:42:10+03:00GitHubnoreply@github.com2025-09-09T08:42:10+03:00contrib : add notes about merging PRs (#15881)
66270cd37dbbebdb7a2f84f08207f18eabb0b291a55acc1b008cfd95e63d5f99a370dbffb98e5a99d2cDaniel Beveniusdaniel.bevenius@gmail.com2025-09-09T06:06:52+02:00GitHubnoreply@github.com2025-09-09T06:06:52+02:00requirements : update transformers/torch for Embedding Gemma (#15828)
663acc1b008cfd95e63d5f99a370dbffb98e5a99d2c7057faf64b514e991e2f70147f82bb13d544b1c0Piotr Wilkin (ilintar)piotr.wilkin@syndatis.com2025-09-09T06:05:55+02:00GitHubnoreply@github.com2025-09-09T06:05:55+02:00model-conversion : add extra debugging support for model conversion (#15877)
6647057faf64b514e991e2f70147f82bb13d544b1c0fe1c92cd7bb491e9c5767c9de413f2b7dc1e832bAldehir Rojashello@alde.dev2025-09-08T16:14:32-05:00GitHubnoreply@github.com2025-09-08T16:14:32-05:00json : support `enum` values within `allOf` (#15830)
665fe1c92cd7bb491e9c5767c9de413f2b7dc1e832be68aa10d8f3d26fdad5b912540362d79de5460e3j-kdev@j-k.io2025-09-08T19:57:01+01:00GitHubnoreply@github.com2025-09-08T21:57:01+03:00media : add llama1 icon (#15878)
666e68aa10d8f3d26fdad5b912540362d79de5460e30a16bf52e6874369cb4fd2bdc4863ef984b688e7Jeff Bolzjbolz@nvidia.com2025-09-08T13:10:07-05:00GitHubnoreply@github.com2025-09-09T02:10:07+08:00vulkan: sort graph to allow more parallel execution (#15850)
6670a16bf52e6874369cb4fd2bdc4863ef984b688e788021565f08e0b7c4e07ac089a15ec16fae9166cAman Guptaamangupta052@gmail.com2025-09-09T01:23:46+08:00GitHubnoreply@github.com2025-09-09T01:23:46+08:00CUDA: generate_cu_files.py - add missing mxfp4 (#15880)
66888021565f08e0b7c4e07ac089a15ec16fae9166c56920f56651908d5cce7a310dabf54ac4f6fbb7fJessejesse@createthis.com2025-09-08T10:59:48-04:00GitHubnoreply@github.com2025-09-08T16:59:48+02:00chat : Deepseek V3.1 reasoning and tool calling support (OpenAI Style) (#15533)
66956920f56651908d5cce7a310dabf54ac4f6fbb7fb0d52998b962bd2681c34bf52af993af79f178b8Xuan-Son Nguyenson@huggingface.co2025-09-08T21:50:05+07:00GitHubnoreply@github.com2025-09-08T16:50:05+02:00server : bring back timings_per_token (#15879)
670b0d52998b962bd2681c34bf52af993af79f178b8f28d4f4ac963f182ea9d0fe9e269f3f5f3782aafGeorgi Gerganovggerganov@gmail.com2025-09-08T13:56:51+03:00GitHubnoreply@github.com2025-09-08T13:56:51+03:00cuda : fix supports_op condition for get_rows when number of blocks is too large (#15868)
671f28d4f4ac963f182ea9d0fe9e269f3f5f3782aaf9fcb29f22f5c33c04c7f0daebb24057899d67a1aGeorgi Gerganovggerganov@gmail.com2025-09-08T13:34:56+03:00GitHubnoreply@github.com2025-09-08T13:34:56+03:00metal : refactor + optimize (#15857)
6729fcb29f22f5c33c04c7f0daebb24057899d67a1a5ef22d281de9c5eaaf616874bc490b89241128cbXuan-Son Nguyenson@huggingface.co2025-09-08T17:33:01+07:00GitHubnoreply@github.com2025-09-08T12:33:01+02:00ggml: allow casting between f32 and i32 (#15783)
6735ef22d281de9c5eaaf616874bc490b89241128cb233d773d02c37982badddf3994e9953a175f34daSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-08T11:55:44+02:00GitHubnoreply@github.com2025-09-08T12:55:44+03:00CUDA: non-contiguous src0 not supported for PAD (#15869)
674233d773d02c37982badddf3994e9953a175f34daa885dcff11a7b73f9377812d6151f6b15d307de0Daniel Beveniusdaniel.bevenius@gmail.com2025-09-08T09:44:34+02:00GitHubnoreply@github.com2025-09-08T09:44:34+02:00convert : force setting sliding_window from original config (#15867)
675a885dcff11a7b73f9377812d6151f6b15d307de0663027fd5490438ce9c27ea866a560e1e268d11fGeorgi Gerganovggerganov@gmail.com2025-09-08T10:27:07+03:00GitHubnoreply@github.com2025-09-08T10:27:07+03:00batched-bench : fix llama_synchronize usage during prompt processing (#15835)
676663027fd5490438ce9c27ea866a560e1e268d11fcf0e3ba1500bd23635b444f64ba23cbdb56c92efGeorgi Gerganovggerganov@gmail.com2025-09-08T10:26:36+03:00GitHubnoreply@github.com2025-09-08T10:26:36+03:00context : fix n_outputs during reserve (#15858)
677cf0e3ba1500bd23635b444f64ba23cbdb56c92efd413dca00360a7e4cb71441dacecfa32556fcc31Georgi Gerganovggerganov@gmail.com2025-09-08T10:25:33+03:00GitHubnoreply@github.com2025-09-08T10:25:33+03:00model : avoid ggml_cont_3d for fused QKV weights (#15662)
678d413dca00360a7e4cb71441dacecfa32556fcc3185ca66a74676e6d5df4433016488e039a4b464aeJeff Bolzjbolz@nvidia.com2025-09-07T23:23:41-05:00GitHubnoreply@github.com2025-09-07T23:23:41-05:00tests: large sizes for get_rows (#15687)
67985ca66a74676e6d5df4433016488e039a4b464ae3976dfbe00f02a62c0deca32c46138e4f0ca81d8Chenguang Li757486878@qq.com2025-09-08T10:03:29+08:00GitHubnoreply@github.com2025-09-08T10:03:29+08:00CANN: Stream sync between devices for acl_graph (#15809)
6803976dfbe00f02a62c0deca32c46138e4f0ca81d8d36e61c580bf7fc7879c443c542312a42b718e11Jeff Bolzjbolz@nvidia.com2025-09-07T13:50:26-05:00GitHubnoreply@github.com2025-09-07T13:50:26-05:00vulkan: support im2col_3d (#15795)
681d36e61c580bf7fc7879c443c542312a42b718e11c97b5e5854b47b18a248d77edb693c63018a0865Aaron Teoaaron.teo1@ibm.com2025-09-08T02:18:28+08:00GitHubnoreply@github.com2025-09-08T02:18:28+08:00ggml-cpu: clean up s390x SIMD (#15855)
682c97b5e5854b47b18a248d77edb693c63018a0865267e99867f09bec8bcc2e424ad9bcddd6cccf9d0Jeff Bolzjbolz@nvidia.com2025-09-07T12:00:49-05:00GitHubnoreply@github.com2025-09-07T19:00:49+02:00vulkan: Support pad_ext (#15794)
683267e99867f09bec8bcc2e424ad9bcddd6cccf9d03b15924d71237a43bb5ad71f5b885ee66a821342Jeff Bolzjbolz@nvidia.com2025-09-07T11:53:07-05:00GitHubnoreply@github.com2025-09-07T18:53:07+02:00vulkan: Use larger loads in scalar/coopmat1 matmul (#15729)
6843b15924d71237a43bb5ad71f5b885ee66a82134279bc429262268ad2ac8a364cfe6c2d6b9c5f008aDaniel Beveniusdaniel.bevenius@gmail.com2025-09-07T10:19:45+02:00GitHubnoreply@github.com2025-09-07T11:19:45+03:00ggml WebGPU: remove userdata from request adapter callback (#15527)
68579bc429262268ad2ac8a364cfe6c2d6b9c5f008ac4df49a42d396bdf7344501813e7de53bc9e7bb3Johannes Gäßlerjohannesg@5d6.de2025-09-07T00:26:28+02:00GitHubnoreply@github.com2025-09-07T00:26:28+02:00CUDA: faster tile FA (Pascal/AMD), headsize 256 (#15769)
686c4df49a42d396bdf7344501813e7de53bc9e7bb33c3635d2f20424d557b5b0605a2a356214ffe048Charles Xucharles.xu@arm.com2025-09-06T16:08:43+02:00GitHubnoreply@github.com2025-09-06T22:08:43+08:00kleidiai: generalize compute_forward_kv_cache to compute_forward_fp16 (#15817)
6873c3635d2f20424d557b5b0605a2a356214ffe04861bdfd5298a78593be649a1035ee2a120b13c4f0Xuan-Son Nguyenson@huggingface.co2025-09-06T19:45:24+07:00GitHubnoreply@github.com2025-09-06T14:45:24+02:00server : speed up tests (#15836)
68861bdfd5298a78593be649a1035ee2a120b13c4f001806e77714ae8a78130d432945b959a0956c56fXuan-Son Nguyenson@huggingface.co2025-09-06T18:35:04+07:00GitHubnoreply@github.com2025-09-06T13:35:04+02:00server : implement prompt processing progress report in stream mode (#15827)
689186415d59552bd5c90549cd8e9be3cc287621fd9fd621880f3fd908424e675a41715a2dc760247a2Aaron Teoaaron.teo1@ibm.com2025-09-06T11:27:28+08:00GitHubnoreply@github.com2025-09-06T11:27:28+08:00ggml-cpu: drop support for nnpa intrinsics (#15821)
690fd621880f3fd908424e675a41715a2dc760247a24281c7b315f8a904549fe527039b976e08098d1aGabe Goodhartghart@us.ibm.com2025-09-05T17:32:39-06:00GitHubnoreply@github.com2025-09-05T17:32:39-06:00aLoRA Support (#15327)
6914281c7b315f8a904549fe527039b976e08098d1a5fac79cbc77b6d12c9feb5f34fc63586b35fd561Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-09-06T01:21:15+02:00GitHubnoreply@github.com2025-09-06T01:21:15+02:00ci : exempt correct research label (#15825)
6925fac79cbc77b6d12c9feb5f34fc63586b35fd561408ff524b40baf4f51a81d42a9828200dd4fcb6bGabe Goodhartghart@us.ibm.com2025-09-05T14:31:24-06:00GitHubnoreply@github.com2025-09-05T14:31:24-06:00Thinking model disabled assistant prefill (#15404)
693408ff524b40baf4f51a81d42a9828200dd4fcb6b5143fa895e7725c5bd2135daf7d8f793d98fa91cEric Curtinericcurtin17@gmail.com2025-09-05T19:43:59+01:00GitHubnoreply@github.com2025-09-05T19:43:59+01:00Implement --log-colors with always/never/auto (#15792)
6945143fa895e7725c5bd2135daf7d8f793d98fa91c3a550b5ca4565c9e28f63880d47840feb27d0ff6Johannes Gäßlerjohannesg@5d6.de2025-09-05T16:07:02+02:00GitHubnoreply@github.com2025-09-05T16:07:02+02:00CUDA: fastdiv, launch bounds for mmvq + q8_1 quant (#15802)
6953a550b5ca4565c9e28f63880d47840feb27d0ff6a81283820a466f2ace06ce4d4bc9512761f9365fDaniel Beveniusdaniel.bevenius@gmail.com2025-09-05T14:49:21+02:00GitHubnoreply@github.com2025-09-05T13:49:21+01:00tests : add --list-ops and --show-coverage options (#15745)
696a81283820a466f2ace06ce4d4bc9512761f9365fc610b6c11b1ef7d678671dcf15acd7187a7ad8f3Erik ScholzGreen-Sky@users.noreply.github.com2025-09-05T11:34:28+02:00GitHubnoreply@github.com2025-09-05T11:34:28+02:00gguf: gguf_writer refactor (#15691)
697c610b6c11b1ef7d678671dcf15acd7187a7ad8f35d6688de08e73acc2532d668380801ed79d704ebGeorgi Gerganovggerganov@gmail.com2025-09-05T10:39:22+03:00GitHubnoreply@github.com2025-09-05T10:39:22+03:00kv-cache : fix SWA checks + disable cacheless iSWA (#15811)
698dcbe998a6d8f13b6c9cb36bbe7b355ced2ac384287932ec016f0ec81e98a13ce5212851f8c12458aYunzhe Jiayunzhe@calculet.tech2025-09-05T14:34:07+08:00Yunzhe Jiayunzhe@calculet.tech2025-09-05T14:34:07+08:00add calrt API used in server side
6995d6688de08e73acc2532d668380801ed79d704eb4fd1242bef6cb2325b4ff1c1a80f3b54b64508a6Daniel Beveniusdaniel.bevenius@gmail.com2025-09-05T04:36:23+02:00GitHubnoreply@github.com2025-09-05T04:36:23+02:00model-conversion : add --embeddings flag to modelcard.template [no ci] (#15801)
7004fd1242bef6cb2325b4ff1c1a80f3b54b64508a6b2426e469e2fdb6c44216d56baa4cfff4f39ae00ExtReMLapin3909752+ExtReMLapin@users.noreply.github.com2025-09-05T01:24:08+02:00GitHubnoreply@github.com2025-09-05T01:24:08+02:00chat : fixed crash when Hermes 2 <tool_call> had a newline before it (#15639)
701b2426e469e2fdb6c44216d56baa4cfff4f39ae009e2b1e83c68a38ea0c64f726dd979439bd02189bPiotr Wilkin (ilintar)piotr.wilkin@syndatis.com2025-09-05T01:22:22+02:00GitHubnoreply@github.com2025-09-05T01:22:22+02:00chat : nemotron thinking & toolcalling support (#15676)
7029e2b1e83c68a38ea0c64f726dd979439bd02189bfb15d649ed14ab447eeab911e0c9d21e35fb243ePiotr Wilkin (ilintar)piotr.wilkin@syndatis.com2025-09-05T01:05:12+02:00GitHubnoreply@github.com2025-09-05T01:05:12+02:00scripts : add Jinja tester PySide6 simple app (#15756)
703fb15d649ed14ab447eeab911e0c9d21e35fb243e856ed0947f27b4ec3ad269fceda0402fbab263d3Daniel Beveniusdaniel.bevenius@gmail.com2025-09-04T18:10:29+02:00GitHubnoreply@github.com2025-09-04T18:10:29+02:00llama : add support for EmbeddingGemma 300m (#15798)
704856ed0947f27b4ec3ad269fceda0402fbab263d3d1e2adba65208d6de3d7f6f23b00c562ff5bb777Gabe Goodhartghart@us.ibm.com2025-09-04T09:53:22-06:00GitHubnoreply@github.com2025-09-04T18:53:22+03:00metal : Add template specialization for mul_mm_id w/ ne20 == 10 (#15799)
705d1e2adba65208d6de3d7f6f23b00c562ff5bb777c1c354e44c06d259679bb5bb7a8fa9f0b28480e4Daniel Beveniusdaniel.bevenius@gmail.com2025-09-04T15:40:44+02:00GitHubnoreply@github.com2025-09-04T15:40:44+02:00llama : set n_outputs to 1 to avoid 0 outputs mean-pooling (#15791)
706c1c354e44c06d259679bb5bb7a8fa9f0b28480e4a68d9144262f1d0ef4f6ba7ad4a7e73e977ba78cChenguang Li757486878@qq.com2025-09-04T20:20:14+08:00GitHubnoreply@github.com2025-09-04T20:20:14+08:00CANN: Refactor ND to NZ workspace to be per-device (#15763)
707a68d9144262f1d0ef4f6ba7ad4a7e73e977ba78cbadb80cadbc40e047b30c43611aba575fc8d6845Xuan-Son Nguyenson@huggingface.co2025-09-04T11:50:23+02:00GitHubnoreply@github.com2025-09-04T11:50:23+02:00server: add exceed_context_size_error type (#15780)
708badb80cadbc40e047b30c43611aba575fc8d68450a1b3982cd0bd18730d50a693053b88c13fd04a6Eric Curtinecurtin@redhat.com2025-09-04T10:49:44+01:00GitHubnoreply@github.com2025-09-04T10:49:44+01:00Document the new max GPU layers default in help (#15771)
7090a1b3982cd0bd18730d50a693053b88c13fd04a65421f63ab08ca1e1a093662a5ccd0117e461185fleejetleejet714@gmail.com2025-09-04T16:38:49+08:00GitHubnoreply@github.com2025-09-04T10:38:49+02:00ggml: add ops for WAN video model (cuda && cpu) (#15669)
7105421f63ab08ca1e1a093662a5ccd0117e461185f820bc9853100708011036e10f40b923968dd9b66hipuddinghuafengchun@gmail.com2025-09-04T15:12:30+08:00GitHubnoreply@github.com2025-09-04T15:12:30+08:00CANN: Fix precision issue on 310I DUO multi-devices (#15784)
711820bc9853100708011036e10f40b923968dd9b66239b60e8986bbcb944f57311a02ac994431bf652rmatifrmatif@proton.me2025-09-04T08:30:28+02:00GitHubnoreply@github.com2025-09-03T23:30:28-07:00opencl: add hs=40 to FA (#15758)
712239b60e8986bbcb944f57311a02ac994431bf652dff7551bfdbdd6e57c13e523d7dcca317640e907Chenguang Li757486878@qq.com2025-09-04T11:03:02+08:00GitHubnoreply@github.com2025-09-04T11:03:02+08:00CANN: fix acl_rstd allocation size in ggml_cann_rms_norm (#15760)
713dff7551bfdbdd6e57c13e523d7dcca317640e9070fce7a1248b74148c1eb0d368b7e18e8bcb96809Ruben Ortlampicard12@live.de2025-09-03T22:55:10+02:00GitHubnoreply@github.com2025-09-03T21:55:10+01:00vulkan: fix mmv subgroup16 selection (#15775)
7140fce7a1248b74148c1eb0d368b7e18e8bcb968098227695d7a3e5b357cb37fad263f00a8ca6db710Jeff Bolzjbolz@nvidia.com2025-09-03T13:33:15-05:00GitHubnoreply@github.com2025-09-03T20:33:15+02:00vulkan: don't use std::string in load_shaders, to improve compile time (#15724)
7158227695d7a3e5b357cb37fad263f00a8ca6db7100014fb4add3a7d000c14b763538045c6170f57d9Daniel Beveniusdaniel.bevenius@gmail.com2025-09-03T20:24:50+02:00GitHubnoreply@github.com2025-09-03T20:24:50+02:00vulkan : update ggml_vk_instance_validation_ext_available (#15666)
7160014fb4add3a7d000c14b763538045c6170f57d9661ae31c9c68201577e70278285b349a5a662cafShin-myoung-serprelent95@naver.com2025-09-04T03:22:55+09:00GitHubnoreply@github.com2025-09-03T20:22:55+02:00ggml vulkan: add hardsigmoid and hardswish operations (#15762)
717661ae31c9c68201577e70278285b349a5a662caf407c23786dd0d3a503e9429eead96e611d3950c9Oliver Simonsosimons@nvidia.com2025-09-03T19:59:16+02:00GitHubnoreply@github.com2025-09-03T19:59:16+02:00CUDA: Optimize `rms_norm_f32` kernel and its fused variants, giving 1-6% perf E2E (#15715)
718407c23786dd0d3a503e9429eead96e611d3950c9cdedb70a998cea7052560fe0b0615a839443564dDaniel Beveniusdaniel.bevenius@gmail.com2025-09-03T18:28:36+02:00GitHubnoreply@github.com2025-09-03T18:28:36+02:00model-conversion : fix pyright errors (#15770)
719cdedb70a998cea7052560fe0b0615a839443564d2c8dac72eb6acd4e20c0da251535dfc46d35178bGeorgi Gerganovggerganov@gmail.com2025-09-03T18:16:26+03:00GitHubnoreply@github.com2025-09-03T18:16:26+03:00sampling : optimize dist sampler (#15704)
7202c8dac72eb6acd4e20c0da251535dfc46d35178b40a751ea9a94364da73537b86502a808ebe1fc3aDaniel Beveniusdaniel.bevenius@gmail.com2025-09-03T13:35:49+02:00GitHubnoreply@github.com2025-09-03T13:35:49+02:00llama : fix incorrect model type for Gemma 270M (#15764)
72140a751ea9a94364da73537b86502a808ebe1fc3a5eae9348835037046f33fafd852df7c952c95209Daniel Beveniusdaniel.bevenius@gmail.com2025-09-03T12:50:47+02:00GitHubnoreply@github.com2025-09-03T12:50:47+02:00model-conversion : remove hardcoded /bin/bash shebangs [no ci] (#15765)
7225eae9348835037046f33fafd852df7c952c9520905c0380f2ae5c7193445e03ebc7b6de9ce49f1a6hipuddinghuafengchun@gmail.com2025-09-03T16:46:01+08:00GitHubnoreply@github.com2025-09-03T16:46:01+08:00CANN: Add RoPE contiguous check for 310I DUP device (#15735)
72305c0380f2ae5c7193445e03ebc7b6de9ce49f1a68c3fdf44ecf08335942f2ba558955f55e88c7991xctanxc-tan@outlook.com2025-09-03T16:16:21+08:00GitHubnoreply@github.com2025-09-03T16:16:21+08:00ggml-cpu : optimize RVV kernels (#15720)
7248c3fdf44ecf08335942f2ba558955f55e88c7991f6da8cb86a28f0319b40d9d2a957a26a7d875f8cDaniel Beveniusdaniel.bevenius@gmail.com2025-09-03T09:48:35+02:00GitHubnoreply@github.com2025-09-03T09:48:35+02:00model-conversion : add missing curl script [no ci] (#15761)
725f6da8cb86a28f0319b40d9d2a957a26a7d875f8c8a2234ea0c89f212190c176d741b7742f0082582hipuddinghuafengchun@gmail.com2025-09-03T14:08:22+08:00GitHubnoreply@github.com2025-09-03T14:08:22+08:00CANN: Mask unsupported TRANSPOSE_1D operator (#15733)
7268a2234ea0c89f212190c176d741b7742f00825823de008208b9b8a33f49f979097a99b4d59e6e521Chenguang Li757486878@qq.com2025-09-03T10:43:53+08:00GitHubnoreply@github.com2025-09-03T10:43:53+08:00CANN: Fix type float_t to float (#15736)
7273de008208b9b8a33f49f979097a99b4d59e6e52169db8a52e6dd0db81ec05c581150cc1e43b8ac46SnA1lGo44647694+skrandy@users.noreply.github.com2025-09-03T03:27:30+08:00GitHubnoreply@github.com2025-09-02T21:27:30+02:00fix: resolve unsigned int initialization warning for n_dims/size in gguf.cpp (#15754)
72869db8a52e6dd0db81ec05c581150cc1e43b8ac46c466abe1587bde458c806e2134e37750f2edf8fbOliver Simonsosimons@nvidia.com2025-09-02T19:40:37+02:00GitHubnoreply@github.com2025-09-03T01:40:37+08:00chore: Update `.clang-format` to use `BinPackArguments=true` (#15744)
729c466abe1587bde458c806e2134e37750f2edf8fb0a2a3841e8ebc570da343e8cf58a21c1010b41e7Johannes Gäßlerjohannesg@5d6.de2025-09-02T18:17:26+02:00GitHubnoreply@github.com2025-09-02T18:17:26+02:00llama: -fa 1/0/-1 aliases for -fa on/off/auto (#15746)
7300a2a3841e8ebc570da343e8cf58a21c1010b41e79961d244f2df6baf40af2f1ddc0927f8d91578c8Ruben Ortlampicard12@live.de2025-09-02T16:02:26+02:00GitHubnoreply@github.com2025-09-02T16:02:26+02:00vulkan: fix shaders gen when no integer dot is available (#15740)
7319961d244f2df6baf40af2f1ddc0927f8d91578c825f1045f07cf0daf667d63e35618842e3174a8c7hipuddinghuafengchun@gmail.com2025-09-02T17:12:37+08:00GitHubnoreply@github.com2025-09-02T17:12:37+08:00CANN: Resolve soft_max precision issue (#15730)
73225f1045f07cf0daf667d63e35618842e3174a8c797669e40735d08db65ef094a8faae8e8411a97dbJeff Bolzjbolz@nvidia.com2025-09-02T01:37:01-05:00GitHubnoreply@github.com2025-09-02T14:37:01+08:00vulkan: Fix macro parameter order for f32 matmul shaders (#15716)
73397669e40735d08db65ef094a8faae8e8411a97db2f853687b3bce15e143a22f678d1715060fd606crmatifrmatif@proton.me2025-09-02T08:26:53+02:00GitHubnoreply@github.com2025-09-01T23:26:53-07:00opencl: add attn sinks support for FA kernels (#15706)
7342f853687b3bce15e143a22f678d1715060fd606cef2af57ddf04ab367f6ba6e8ed100fc56785e4b7Chenguang Li757486878@qq.com2025-09-02T14:07:48+08:00GitHubnoreply@github.com2025-09-02T14:07:48+08:00CANN: Support eager execution mode under ACL graph compilation (#15712)
735ef2af57ddf04ab367f6ba6e8ed100fc56785e4b75d804a4938d896f9089687a8b99911cdb43b0b94hipuddinghuafengchun@gmail.com2025-09-02T14:05:23+08:00GitHubnoreply@github.com2025-09-02T14:05:23+08:00CANN: Support ext_factor in rope (#15710)
7365d804a4938d896f9089687a8b99911cdb43b0b94d4d8dbe383e8b9600cbe8b42016e3a4529b51219Johannes Gäßlerjohannesg@5d6.de2025-09-02T01:14:55+02:00GitHubnoreply@github.com2025-09-01T16:14:55-07:00ggml-backend: raise GGML_MAX_SPLIT_INPUTS (#15722)
737d4d8dbe383e8b9600cbe8b42016e3a4529b5121935a42edac84690990a75726898d975cd08d220c0Gilad S.7817232+giladgd@users.noreply.github.com2025-09-01T22:17:42+03:00GitHubnoreply@github.com2025-09-01T21:17:42+02:00vulkan: use memory budget extension to read memory usage (#15545)
73835a42edac84690990a75726898d975cd08d220c0fec7911f8febb22c27ce1acb12441800da24cc06Jeff Bolzjbolz@nvidia.com2025-09-01T14:01:10-05:00GitHubnoreply@github.com2025-09-01T21:01:10+02:00vulkan: add missing clamps in new mul_mat_id paths (#15702)
739fec7911f8febb22c27ce1acb12441800da24cc06078ce23ea77988f2fe1a42afb18d58bc084d55faRuben Ortlampicard12@live.de2025-09-01T20:58:35+02:00GitHubnoreply@github.com2025-09-01T20:58:35+02:00vulkan: disable large mmv subgroups on older Nvidia GPUs (#15717)
740078ce23ea77988f2fe1a42afb18d58bc084d55faa0c2b207c596d1092a08615de61ab56a7d63515fs-goto-11206795233+s-goto-11@users.noreply.github.com2025-09-02T03:13:49+09:00GitHubnoreply@github.com2025-09-01T20:13:49+02:00ggml: SVE support for exponential functions (#15145)
741a0c2b207c596d1092a08615de61ab56a7d63515f4b20d8b7e31718d3fe8b4f12d220a9d19e4f1997Prashant Vithule119530321+Vithulep@users.noreply.github.com2025-09-01T23:43:16+05:30GitHubnoreply@github.com2025-09-01T20:13:16+02:00ggml: aarch64: Implement SVE F16 kernels for vector functions (#15115)
7424b20d8b7e31718d3fe8b4f12d220a9d19e4f199702c1813517412f3e00aa6ca7c0273fea64edb492Jie Fu (傅杰)jiefu@tencent.com2025-09-01T23:53:31+08:00GitHubnoreply@github.com2025-09-01T23:53:31+08:00convert : remove redundant code (#15708)
74302c1813517412f3e00aa6ca7c0273fea64edb49277dee9de97be75b7143a213bc48893e0c0b29af7Ruben Ortlampicard12@live.de2025-09-01T16:19:07+02:00GitHubnoreply@github.com2025-09-01T16:19:07+02:00Vulkan: Add Integer Dot Product mul_mat_vec shader for legacy quants (#14903)
74477dee9de97be75b7143a213bc48893e0c0b29af74795c91c32fec7165a1364763d4d4f0c93abf933Daniel Beveniusdaniel.bevenius@gmail.com2025-09-01T14:28:49+02:00GitHubnoreply@github.com2025-09-01T14:28:49+02:00ggml : WebGPU add TRANSPOSE and RESHAPE to supported ops (#15695)
7454795c91c32fec7165a1364763d4d4f0c93abf933b66df9d9c942254d03209186ef24ed7c994a576eJie Fu (傅杰)jiefu@tencent.com2025-09-01T15:34:59+08:00GitHubnoreply@github.com2025-09-01T10:34:59+03:00docs : add Hunyuan to models section (#15707)
746b66df9d9c942254d03209186ef24ed7c994a576eb9382c3877c6067feccf182efe9449a2d1cb24c7Akarshan Biswasakarshan@menlo.ai2025-09-01T06:55:06+05:30GitHubnoreply@github.com2025-09-01T06:55:06+05:30CUDA: fix build error from ambiguous __half conversions in conv2d (#15690)
747b9382c3877c6067feccf182efe9449a2d1cb24c73dc7397a2799bdc07bccf637ab7ae5a1e786d1a4hipuddinghuafengchun@gmail.com2025-09-01T08:57:23+08:00GitHubnoreply@github.com2025-09-01T08:57:23+08:00CANN: Optimize MUL_MAT_ID (#15658)
7483dc7397a2799bdc07bccf637ab7ae5a1e786d1a4e92d53b29e393fc4c0f9f1f7c3fe651be8d36faahipuddinghuafengchun@gmail.com2025-09-01T08:57:00+08:00GitHubnoreply@github.com2025-09-01T08:57:00+08:00CANN: fix RoPE cache issue on multi-device (#15629)
749e92d53b29e393fc4c0f9f1f7c3fe651be8d36faa0d161f021aa33ec0e90cce96f5d1a88925557327Georgi Gerganovggerganov@gmail.com2025-08-31T20:41:02+03:00GitHubnoreply@github.com2025-08-31T20:41:02+03:00sampling : optimize samplers by reusing bucket sort (#15665)
7500d161f021aa33ec0e90cce96f5d1a889255573274efd5a83163ff383285b3a4c2106feabf5c69557Georgi Gerganovggerganov@gmail.com2025-08-31T20:11:58+03:00GitHubnoreply@github.com2025-08-31T20:11:58+03:00server : enable /slots by default and make it secure (#15630)
7514efd5a83163ff383285b3a4c2106feabf5c69557274966226f87f301ac132da898280ca3142b60e5Georgi Gerganovggerganov@gmail.com2025-08-31T19:43:30+03:00GitHubnoreply@github.com2025-08-31T19:43:30+03:00metal : fix checks for available FA kernels (#15700)
752274966226f87f301ac132da898280ca3142b60e59777032dccd67bdc7785aeab7497014a8be8daccDiego Devesaslarengh@gmail.com2025-08-31T08:47:05-07:00GitHubnoreply@github.com2025-08-31T18:47:05+03:00llama : fix fattn reserve call n_seqs parameter (#15699)
7539777032dccd67bdc7785aeab7497014a8be8dacc7d3c9f2b217acf0ce5db81ae83d3f375f49ab2c7Diego Devesaslarengh@gmail.com2025-08-31T06:49:03-07:00GitHubnoreply@github.com2025-08-31T15:49:03+02:00llama : separate compute buffer reserve from fattn check (#15696)
7547d3c9f2b217acf0ce5db81ae83d3f375f49ab2c7bbbf5ecccb35286521f735239d499eec4279a840Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-31T15:30:20+02:00GitHubnoreply@github.com2025-08-31T15:30:20+02:00ci : explicitly set fa off or on (#15692)
755bbbf5ecccb35286521f735239d499eec4279a840c37052ab4d6d1ae73c0e90bc6e560cc6409e1311Jeff Bolzjbolz@nvidia.com2025-08-31T03:13:27-05:00GitHubnoreply@github.com2025-08-31T10:13:27+02:00vulkan: handle large sizes for get_rows (#15686)
756c37052ab4d6d1ae73c0e90bc6e560cc6409e13115c16b9c87d840e4d5d55fa83c732c6b693346f40Jeff Bolzjbolz@nvidia.com2025-08-31T02:06:43-05:00GitHubnoreply@github.com2025-08-31T09:06:43+02:00vulkan: mul_mat_id coopmat2 optimizations (#15546)
7575c16b9c87d840e4d5d55fa83c732c6b693346f40b97c9edc59d4a1b4069991aa670411190f4f3a3eDaniel Beveniusdaniel.bevenius@gmail.com2025-08-31T08:46:42+02:00GitHubnoreply@github.com2025-08-31T08:46:42+02:00vulkan : remove unused portability_enumeration_ext variable (#15679)
758b97c9edc59d4a1b4069991aa670411190f4f3a3e94e82c7eadeb8fff0db4bfd1ab6d8cf65fa6f2e0Jeff Bolzjbolz@nvidia.com2025-08-31T01:30:54-05:00GitHubnoreply@github.com2025-08-31T08:30:54+02:00vulkan: Allow fallback to sysmem memory when vidmem is full (#15649)
75994e82c7eadeb8fff0db4bfd1ab6d8cf65fa6f2e04d74393bcc956ccd7df68a6a06d1a0575cfa712cJeff Bolzjbolz@nvidia.com2025-08-31T01:27:57-05:00GitHubnoreply@github.com2025-08-31T08:27:57+02:00vulkan: clamp matmul and FA results to the max finite value (#15652)
7604d74393bcc956ccd7df68a6a06d1a0575cfa712cdd892555b0681b7f56d38780f6fdfe00a195160fCharles Xucharles.xu@arm.com2025-08-30T18:03:42+02:00GitHubnoreply@github.com2025-08-31T00:03:42+08:00ggml: update kleidiai to v1.13.0 (#15663)
761dd892555b0681b7f56d38780f6fdfe00a195160fe81b8e4b7f5ab870836fad26d154a7507b341b36Diego Devesaslarengh@gmail.com2025-08-30T08:51:28-07:00GitHubnoreply@github.com2025-08-30T23:51:28+08:00Update build.md to remove MSVC arm64 notes (#15684)
762e81b8e4b7f5ab870836fad26d154a7507b341b3638ad381f9f5d4dd368a96d844fb19cf501ed9d22Johannes Gäßlerjohannesg@5d6.de2025-08-30T16:32:10+02:00GitHubnoreply@github.com2025-08-30T16:32:10+02:00llama: use FA + max. GPU layers by default (#15434)
76338ad381f9f5d4dd368a96d844fb19cf501ed9d22696fccf354e9dbdfbce135bc40b44c9dcc64dda9Johannes Gäßlerjohannesg@5d6.de2025-08-30T16:20:32+02:00GitHubnoreply@github.com2025-08-30T16:20:32+02:00CUDA: use FP32 arithmetic for conv2d (#15683)
764696fccf354e9dbdfbce135bc40b44c9dcc64dda9ef476916bba4b44f44be0c98babc1cb025968e75Jeff Bolzjbolz@nvidia.com2025-08-30T04:11:22-05:00GitHubnoreply@github.com2025-08-30T11:11:22+02:00vulkan: Skip syncing for prealloc_y when it is reused (#15544)
765ef476916bba4b44f44be0c98babc1cb025968e75d82f6aa34a216f5df1945cdfe121ba5e6cd80be0Chenguang Li757486878@qq.com2025-08-30T10:18:35+08:00GitHubnoreply@github.com2025-08-30T10:18:35+08:00CANN: FIx compiler warnings (#15661)
766d82f6aa34a216f5df1945cdfe121ba5e6cd80be03d16b29c3bb1ec816ac0e782f20d169097063919Sergey Alirzaevl29ah@riseup.net2025-08-30T00:12:53+02:00GitHubnoreply@github.com2025-08-30T00:12:53+02:00server : removed obsolete doc (#15670)
767792b44f2ed9668cce7f267ff0ae4950ed9b4a5de81017865ee444cf49ce0136f2be1e41a0270ff91ExtReMLapin3909752+ExtReMLapin@users.noreply.github.com2025-08-29T19:25:40+02:00GitHubnoreply@github.com2025-08-29T20:25:40+03:00server : add documentation for `parallel_tool_calls` param (#15647)
76881017865ee444cf49ce0136f2be1e41a0270ff9160e5eee31f1af9bb579ac45380e3857d610020b9Aman Guptaamangupta052@gmail.com2025-08-29T21:30:06+08:00GitHubnoreply@github.com2025-08-29T21:30:06+08:00CUDA: fix bug in rms_norm fusion (#15660)
76960e5eee31f1af9bb579ac45380e3857d610020b9009b709d6efd24820ac67765ed339a72dc797814Piotr Wilkin (ilintar)piotr.wilkin@syndatis.com2025-08-29T14:53:41+02:00GitHubnoreply@github.com2025-08-29T14:53:41+02:00chat : Seed OSS thinking + tool call support (#15552)
770009b709d6efd24820ac67765ed339a72dc797814e8d99dd0b67f2ecc1e45fca8074a3a18c3e036d2Aman Guptaamangupta052@gmail.com2025-08-29T11:35:58+08:00GitHubnoreply@github.com2025-08-29T11:35:58+08:00CUDA: fuse adds, fuse add with rms norm (#15631)
771e8d99dd0b67f2ecc1e45fca8074a3a18c3e036d2a8bca68f727844e7dcf24a956003b3c2039ea563Gabe Goodhartghart@us.ibm.com2025-08-28T18:39:31-06:00GitHubnoreply@github.com2025-08-28T18:39:31-06:00nvidia nemotron nano v2 (nemotronh) (#15507)
772a8bca68f727844e7dcf24a956003b3c2039ea563c97dc093912ad014f6d22743ede0d4d7fd82365aGabe Goodhartghart@us.ibm.com2025-08-28T15:27:36-05:00GitHubnoreply@github.com2025-08-28T15:27:36-05:00fix: Compute the full sum in llama-eval-callback, not just the sum of printed values (#15637)
773c97dc093912ad014f6d22743ede0d4d7fd82365a6c442f42ff25564a0cd6b1435d9abc1b0178eac5mnehete3233429707+mnehete32@users.noreply.github.com2025-08-29T00:03:03+05:30GitHubnoreply@github.com2025-08-28T20:33:03+02:00CUDA: add conv2d (#15635)
7746c442f42ff25564a0cd6b1435d9abc1b0178eac573804145ab6c06888e314046abf08f66a0133679Aaron Teoaaron.teo1@ibm.com2025-08-28T22:39:27+08:00GitHubnoreply@github.com2025-08-28T22:39:27+08:00ggml-cpu: fix invalid hsum build in debug s390x (#15634)
77573804145ab6c06888e314046abf08f66a0133679c8d0d14e77c3c45df5cbbddde9c2b4c377ec4e7acompiladegit@compilade.net2025-08-28T10:11:36-04:00GitHubnoreply@github.com2025-08-28T10:11:36-04:00ggml : fix SSM_SCAN for n_groups > 1 (#15625)
776c8d0d14e77c3c45df5cbbddde9c2b4c377ec4e7a84ab83cc0b4b7e769451ee48e4c7d1acef91ef25Georgi Gerganovggerganov@gmail.com2025-08-28T17:09:05+03:00GitHubnoreply@github.com2025-08-28T17:09:05+03:00kv-cache : fix find_slot to not search for continuous slot (#15638)
77784ab83cc0b4b7e769451ee48e4c7d1acef91ef2555042b3692cb1467c9ee15c62c4a9fbf180f89e3Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-28T15:49:50+02:00GitHubnoreply@github.com2025-08-28T15:49:50+02:00model : jina-embeddings-v3 support (#13693)
77855042b3692cb1467c9ee15c62c4a9fbf180f89e38a4280ce431da6b33e5a95ae1fd61472c8c3f8ccAman Guptaamangupta052@gmail.com2025-08-28T19:23:22+08:00GitHubnoreply@github.com2025-08-28T19:23:22+08:00scripts: add sqlite3 check for compare-commits.sh (#15633)
7798a4280ce431da6b33e5a95ae1fd61472c8c3f8cc64387f6e95434b393ac3df285864692b7fd9c4d2Georgi Gerganovggerganov@gmail.com2025-08-28T12:27:02+03:00GitHubnoreply@github.com2025-08-28T12:27:02+03:00kv-cache : remove LLAMA_SET_ROWS checks (#15505)
78064387f6e95434b393ac3df285864692b7fd9c4d2d35a1e8c41f747548775225973a99507896a8c61Aleksei Nikiforov103434461+AlekseiNikiforovIBM@users.noreply.github.com2025-08-28T10:56:41+02:00GitHubnoreply@github.com2025-08-28T16:56:41+08:00gguf-py: byteswapping improvements (#12851)
781d35a1e8c41f747548775225973a99507896a8c6146d9caa27a0281150e8cf082308c0f9e7576ebe5Joshua Cogliatijrincayc@users.noreply.github.com2025-08-28T01:48:20-06:00GitHubnoreply@github.com2025-08-28T10:48:20+03:00cli : change log to warning to explain reason for stopping (#15604)
78246d9caa27a0281150e8cf082308c0f9e7576ebe55a0e3ef6f00c658fbae53797f02d5a360ebf8fecDaniel Beveniusdaniel.bevenius@gmail.com2025-08-28T09:26:48+02:00GitHubnoreply@github.com2025-08-28T09:26:48+02:00model-conversion : add mmproj conversion target (#15628)
7835a0e3ef6f00c658fbae53797f02d5a360ebf8fecfbef0fad7a7c765939f6c9e322fa05cd52cf0c15matiaslin45382001+matiaslin@users.noreply.github.com2025-08-27T17:32:36-07:00GitHubnoreply@github.com2025-08-28T02:32:36+02:00cuda: Add cublasLt_static linking when GGML_STATIC is enabled (#15622)
784fbef0fad7a7c765939f6c9e322fa05cd52cf0c15da54f9f1a2db07aaae390024ac466e7867685d94Johannes Gäßlerjohannesg@5d6.de2025-08-27T20:58:09+02:00GitHubnoreply@github.com2025-08-27T20:58:09+02:00server: higher timeout for tests (#15621)
785da54f9f1a2db07aaae390024ac466e7867685d9447373271f971aa5a0a6462b286f4a7b5bd4ba644Georgi Gerganovggerganov@gmail.com2025-08-27T15:48:07+03:00GitHubnoreply@github.com2025-08-27T15:48:07+03:00presets : add qwen3-30B-a3b FIM (#15616)
78647373271f971aa5a0a6462b286f4a7b5bd4ba6441bded5a3b3376b2aa3ba2f11ade5910c550f354buvoscarl@uvos.xyz2025-08-27T13:58:54+02:00GitHubnoreply@github.com2025-08-27T13:58:54+02:00HIP: Enable support for ggml_backend_cuda_register_host_buffer (#15615)
7871bded5a3b3376b2aa3ba2f11ade5910c550f354b1e7489745a74996fc36e8fd05b73aa16bc184e0cGeorgi Gerganovggerganov@gmail.com2025-08-27T13:55:12+03:00GitHubnoreply@github.com2025-08-27T13:55:12+03:00kv-cache : better estimate of n_kv for multi-sequence batches (#15610)
7881e7489745a74996fc36e8fd05b73aa16bc184e0c1cf123a343ab7ca5586aacb9e0a1d2de7fe33be4Chenguang Li757486878@qq.com2025-08-27T17:21:41+08:00GitHubnoreply@github.com2025-08-27T17:21:41+08:00CANN: refactor mask handling and improve performance in FA (#15561)
7891cf123a343ab7ca5586aacb9e0a1d2de7fe33be4fcca2182a18c786c57d02d1d1927204b133f1fdcxctanxc-tan@outlook.com2025-08-27T16:44:22+08:00GitHubnoreply@github.com2025-08-27T16:44:22+08:00ggml-cpu : add basic RVV support for vector f32 ops (#15057)
790fcca2182a18c786c57d02d1d1927204b133f1fdc86076f92de1a547ef87c304facca3b4b9fae6c21Daniel Beveniusdaniel.bevenius@gmail.com2025-08-27T10:28:53+02:00GitHubnoreply@github.com2025-08-27T10:28:53+02:00common : add -m to bash completion for --model [no ci] (#15591)
79186076f92de1a547ef87c304facca3b4b9fae6c21bcbddcd54f0d5c22eab180831fdea6484107112frmatifrmatif@proton.me2025-08-27T08:36:05+02:00GitHubnoreply@github.com2025-08-26T23:36:05-07:00OpenCL: add fused group_norm/norm, mul, add (#15314)
79287932ec016f0ec81e98a13ce5212851f8c12458a0115cfb7c1dec0753aecb34f3af4338bf9de8eebYunzhe Jiayunzhe@calculet.tech2025-08-27T14:29:18+08:00Yunzhe Jiayunzhe@calculet.tech2025-08-27T14:29:18+08:00fix merge problem
793bcbddcd54f0d5c22eab180831fdea6484107112f8b696861364360770e9f61a3422d32941a477824Diego Devesaslarengh@gmail.com2025-08-26T13:14:38-07:00GitHubnoreply@github.com2025-08-26T22:14:38+02:00tests : fix test-opt with GGML_BACKEND_DL (#15599)
7948b696861364360770e9f61a3422d32941a4778248ce3ff1d91245e158d98d8062cd64b0dd98dcfe3Akarshan Biswasakarshan@menlo.ai2025-08-27T00:27:49+05:30GitHubnoreply@github.com2025-08-27T00:27:49+05:30SYCL: fix rms_norm_mul_add for tensor dim not a multiple of sg_size (#15592)
7958ce3ff1d91245e158d98d8062cd64b0dd98dcfe344b1efa41acd4df3f56ee0e46f898135ecd1a054fidoriel49869342+fidoriel@users.noreply.github.com2025-08-26T20:05:50+02:00GitHubnoreply@github.com2025-08-26T20:05:50+02:00mtmd : fix mtmd ios build (#15579)
79644b1efa41acd4df3f56ee0e46f898135ecd1a054a6a58d64785cb458ed9de52f391aa38142d38d64Eve139727413+netrunnereve@users.noreply.github.com2025-08-26T15:42:49ZGitHubnoreply@github.com2025-08-26T15:42:49Ztests: add performance test for mul mat id (#15543)
797a6a58d64785cb458ed9de52f391aa38142d38d640373486dbc0dccbdcb3b5fdd65759d88cec06196shalinib-ibmShalini.Salomi.Bodapati@ibm.com2025-08-26T21:05:25+05:30GitHubnoreply@github.com2025-08-26T23:35:25+08:00llamafile: PowerPC Sgemm Optimization (#15558)
7980373486dbc0dccbdcb3b5fdd65759d88cec0619662cef26ac5b6b7acb635d3dc963813b43952dc2bGeorgi Gerganovggerganov@gmail.com2025-08-26T17:45:17+03:00GitHubnoreply@github.com2025-08-26T17:45:17+03:00graph : fix assert in memory-less build_attn (#15590)
79962cef26ac5b6b7acb635d3dc963813b43952dc2b8f5afa94c4f929da71f560db7c9f38ef6a783d95Daniel Beveniusdaniel.bevenius@gmail.com2025-08-26T16:12:29+02:00GitHubnoreply@github.com2025-08-26T16:12:29+02:00model-conversion : add qat-q4 quantization targets (#15588)
8008f5afa94c4f929da71f560db7c9f38ef6a783d95b3964c1e890ef8c947afb36a5124ce6fcb2136d4Johannes Gäßlerjohannesg@5d6.de2025-08-26T16:01:20+02:00GitHubnoreply@github.com2025-08-26T16:01:20+02:00CUDA: return -1 for nonexistent compiled arch (#15587)
801b3964c1e890ef8c947afb36a5124ce6fcb2136d479a546220c719e6a70627b243a478ab8d84dc9e1Georgi Gerganovggerganov@gmail.com2025-08-26T14:22:14+03:00GitHubnoreply@github.com2025-08-26T14:22:14+03:00metal : optimize FA vec for large sequences and BS <= 8 (#15566)
80279a546220c719e6a70627b243a478ab8d84dc9e185cc1ae998e4898d9fa992cb9b8620338cee97bfXuan-Son Nguyenson@huggingface.co2025-08-26T12:54:19+02:00GitHubnoreply@github.com2025-08-26T12:54:19+02:00mtmd : support Kimi VL model (#15458)
80385cc1ae998e4898d9fa992cb9b8620338cee97bf1d8d83deaa48d4a5491820d58ff1c0d8cf9d196cGeorgi Gerganovggerganov@gmail.com2025-08-26T12:47:00+03:00GitHubnoreply@github.com2025-08-26T12:47:00+03:00context : print graph stats for memory-less contexts (#15586)
8041d8d83deaa48d4a5491820d58ff1c0d8cf9d196cc4e9239064a564de7b94ee2b401ae907235a8fcaGeorgi Gerganovggerganov@gmail.com2025-08-26T12:46:15+03:00GitHubnoreply@github.com2025-08-26T12:46:15+03:00metal : improve `MUL_MAT_ID` (#15541)
805c4e9239064a564de7b94ee2b401ae907235a8fca39842a7f73012eb42816ca4f26411782bd3da7c5tc-mb157115220+tc-mb@users.noreply.github.com2025-08-26T16:05:55+08:00GitHubnoreply@github.com2025-08-26T10:05:55+02:00model : support MiniCPM-V 4.5 (#15575)
8060115cfb7c1dec0753aecb34f3af4338bf9de8eeb3a617cb18121a85eed17ac5585788810c5d9c1e0 0d5a470223fc90b6b6807921d68011ff06ae7f9eYunzhe Jiayunzhe@calculet.tech2025-08-26T15:19:56+08:00Yunzhe Jiayunzhe@calculet.tech2025-08-26T15:19:56+08:00Merge branch 'master' into dev
80739842a7f73012eb42816ca4f26411782bd3da7c50fd90db5858e325358a5fbdaa4e327ee2a79d0d4Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-26T09:08:08+02:00GitHubnoreply@github.com2025-08-26T09:08:08+02:00gguf-py : remove erroneous FFN_GATE entry (#15583)
8080fd90db5858e325358a5fbdaa4e327ee2a79d0d44c37636b3ea96f2574eeb7668b93fcc0e64b05ddSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-26T08:51:43+02:00GitHubnoreply@github.com2025-08-26T09:51:43+03:00metal : remove contiguous assertion for src0 in IM2COL (#15577)
8094c37636b3ea96f2574eeb7668b93fcc0e64b05dd34bdbbd7c2b70b848718e95bc48010f6aecd2816Yoshi_likes_e4104140648+pt13762104@users.noreply.github.com2025-08-26T13:15:33+07:00GitHubnoreply@github.com2025-08-26T08:15:33+02:00Add a warning for special devices (#15563)
81034bdbbd7c2b70b848718e95bc48010f6aecd281674f52f77f28a5ad6d6075231afcb8d1ad763ca32Jeff Bolzjbolz@nvidia.com2025-08-25T23:42:44-05:00GitHubnoreply@github.com2025-08-26T06:42:44+02:00vulkan: Remove splitting for mul_mat_id (#15568)
81174f52f77f28a5ad6d6075231afcb8d1ad763ca32f7207b0415986dd7f48447149da7de3a82338276Qeeweew68716978+Qeeweew@users.noreply.github.com2025-08-26T05:21:22+08:00GitHubnoreply@github.com2025-08-25T23:21:22+02:00CUDA: Accelerate MXFP4 table lookup using `__byte_perm` (#15451)
812f7207b0415986dd7f48447149da7de3a823382764d917cd4f64cc37744e76d084659475819fb0728lhezlih@qti.qualcomm.com2025-08-25T14:18:09-07:00GitHubnoreply@github.com2025-08-25T14:18:09-07:00opencl: fix support ops condition for `rms_norm` (#15560)
8134d917cd4f64cc37744e76d084659475819fb0728886b97a5d693550c2da470c091d9d27bf38398f8Ruben Ortlampicard12@live.de2025-08-25T17:56:59+02:00GitHubnoreply@github.com2025-08-25T17:56:59+02:00vulkan: fix min subgroup 16 condition for mmid subgroup optimization (#15565)
814886b97a5d693550c2da470c091d9d27bf38398f8111f8d06f0b9169059779a35f655b343242ccbb6Jeff Bolzjbolz@nvidia.com2025-08-25T10:47:16-05:00GitHubnoreply@github.com2025-08-25T10:47:16-05:00tests: Generate unique input values for count_equal (#15487)
815111f8d06f0b9169059779a35f655b343242ccbb65eff6ec9b1220b599a43b594b1110487ab6aca08Ihar Hrachyshkaihar.hrachyshka@gmail.com2025-08-25T11:27:34-04:00GitHubnoreply@github.com2025-08-25T18:27:34+03:00metal: fix regression when no metal devices are present (#15531)
8165eff6ec9b1220b599a43b594b1110487ab6aca08dfd9b5f6c7586c88588f06a644c131bec071a0a1Johannes Gäßlerjohannesg@5d6.de2025-08-25T17:23:40+02:00GitHubnoreply@github.com2025-08-25T17:23:40+02:00CUDA: MoE helper in device code, better tile sizes (#15525)
817dfd9b5f6c7586c88588f06a644c131bec071a0a15a6bc6b1a6cb665a944426c2055794950e524bf5Daniel Beveniusdaniel.bevenius@gmail.com2025-08-25T15:00:43+02:00GitHubnoreply@github.com2025-08-25T15:00:43+02:00model-conversion : set pooling type to none in logits.cpp (#15564)
8185a6bc6b1a6cb665a944426c2055794950e524bf56b64f74b55628e4193f4fb00313f07dbd8556528Daniel Beveniusdaniel.bevenius@gmail.com2025-08-25T14:25:25+02:00GitHubnoreply@github.com2025-08-25T14:25:25+02:00model-conversion : add model card template for embeddings [no ci] (#15557)
8196b64f74b55628e4193f4fb00313f07dbd85565280d5a470223fc90b6b6807921d68011ff06ae7f9eGeorgi Gerganovggerganov@gmail.com2025-08-25T13:56:43+03:00GitHubnoreply@github.com2025-08-25T13:56:43+03:00batched-bench : fix unified KV cache handling + pp timing (#15562)
8200d5a470223fc90b6b6807921d68011ff06ae7f9eb0ba31f525a2ad7fb539660be0c8f088c9869f6aWeizhao Ouyango451686892@gmail.com2025-08-25T17:15:06+08:00GitHubnoreply@github.com2025-08-25T11:15:06+02:00origin/master, origin/HEAD, masterconvert : update Ernie 4.5 dense architecture name (#15555)
821b0ba31f525a2ad7fb539660be0c8f088c9869f6a7da9fed0d6f1750a8783436f6f313b87e76e6378Georgi Gerganovggerganov@gmail.com2025-08-25T10:14:48+03:00GitHubnoreply@github.com2025-08-25T10:14:48+03:00metal : add FA kernels for HS=40 (#15559)
8227da9fed0d6f1750a8783436f6f313b87e76e6378c247d06f38fc09059c9607a28aa44f5ff6be208dRunningLeonmnsheng@yeah.net2025-08-25T14:32:16+08:00GitHubnoreply@github.com2025-08-25T08:32:16+02:00convert : support interns1-mini (#15412)
823c247d06f38fc09059c9607a28aa44f5ff6be208d043fb27d3808766d8ea8195bbd12359727264402Chenguang Li757486878@qq.com2025-08-25T10:32:21+08:00GitHubnoreply@github.com2025-08-25T10:32:21+08:00CANN: ROPE cache sin/cos repeat (#15501)
824043fb27d3808766d8ea8195bbd12359727264402b730706a49e576fb882dc34d9966345778b3ab0bRuben Ortlampicard12@live.de2025-08-24T19:36:36+02:00GitHubnoreply@github.com2025-08-24T19:36:36+02:00vulkan: apply MUL_MAT_ID subgroup optimization to non-coopmat devices (#15524)
825b730706a49e576fb882dc34d9966345778b3ab0bc9a24fb93208fbbd3da6d903eb75431bfa97e59eGeorgi Gerganovggerganov@gmail.com2025-08-24T13:07:07+03:00GitHubnoreply@github.com2025-08-24T13:07:07+03:00kv-cache : support layer reuse (#15504)
826c9a24fb93208fbbd3da6d903eb75431bfa97e59ea9c6ffcbfacee092bfaaa400306fceda18199737Jeff Bolzjbolz@nvidia.com2025-08-24T04:24:25-05:00GitHubnoreply@github.com2025-08-24T11:24:25+02:00vulkan: Support FA with any multiple of 8 head sizes (#15537)
827a9c6ffcbfacee092bfaaa400306fceda18199737e78cf0d4b1bdbbc2479f11d58ce0c8f51f755875Ruben Ortlampicard12@live.de2025-08-24T10:48:53+02:00GitHubnoreply@github.com2025-08-24T10:48:53+02:00vulkan: enable Conv2D for Apple after MoltenVK fixed the bug (#15526)
828e78cf0d4b1bdbbc2479f11d58ce0c8f51f755875710dfc465a68f7443b87d9f792cffba00ed739feJeff Bolzjbolz@nvidia.com2025-08-24T03:48:21-05:00GitHubnoreply@github.com2025-08-24T10:48:21+02:00vulkan: workaround MoltenVK compile failure in multi_add (#15506)
829710dfc465a68f7443b87d9f792cffba00ed739fe611f419cff11e4952228162a1c44cb35dff2274aJohannes Gäßlerjohannesg@5d6.de2025-08-23T21:37:06+02:00GitHubnoreply@github.com2025-08-23T21:37:06+02:00CUDA: fix half2 -> half conversion for HIP (#15529)
830611f419cff11e4952228162a1c44cb35dff2274ab1afcab804e3281867a5471fbd701e32eb32e512Jeff Bolzjbolz@nvidia.com2025-08-23T13:16:17-05:00GitHubnoreply@github.com2025-08-23T13:16:17-05:00vulkan: optimize rms_norm, and allow the work to spread across multiple SMs (#15281)
831b1afcab804e3281867a5471fbd701e32eb32e5129ef536907de1b50c30e0369284898d30472a755aPiotr Wilkin (ilintar)piotr.wilkin@syndatis.com2025-08-23T15:21:52+02:00GitHubnoreply@github.com2025-08-23T15:21:52+02:00model : add support for Seed-OSS (#15490)
8329ef536907de1b50c30e0369284898d30472a755a21dc4ddaf21b8ed551d717e7606abd2cffbacdbfJohannes Gäßlerjohannesg@5d6.de2025-08-23T12:58:58+02:00GitHubnoreply@github.com2025-08-23T13:58:58+03:00scripts: fix compare-llama-bench.py (#15521)
83321dc4ddaf21b8ed551d717e7606abd2cffbacdbf289bf4113ef5c02d8f5eb0cf2d86683d8b8bc4d9LaffeyNyaa112215776+LaffeyNyaa@users.noreply.github.com2025-08-23T16:38:30+08:00GitHubnoreply@github.com2025-08-23T10:38:30+02:00chat : fix debug build assertion in trim function (#15520)
834289bf4113ef5c02d8f5eb0cf2d86683d8b8bc4d9b55f06e1aa67fb10e89f53e31bbccf37eb2678eaJeff Bolzjbolz@nvidia.com2025-08-23T02:33:36-05:00GitHubnoreply@github.com2025-08-23T09:33:36+02:00vulkan: Rewrite synchronization to allow some overlap between nodes (#15489)
835b55f06e1aa67fb10e89f53e31bbccf37eb2678ea0a9b43e507a359ca392c037cf341f55137ad0b69R0CKSTARyeahdongcn@gmail.com2025-08-23T14:58:57+08:00GitHubnoreply@github.com2025-08-23T08:58:57+02:00vulkan.Dockerfile: install vulkan SDK using tarball (#15282)
8360a9b43e507a359ca392c037cf341f55137ad0b69330c3d2d21b55bca5517db7d2eea2ea8f131df4aAclyaclysia@gmail.com2025-08-23T08:35:21+02:00GitHubnoreply@github.com2025-08-23T08:35:21+02:00vulkan : support ggml_mean (#15393)
837330c3d2d21b55bca5517db7d2eea2ea8f131df4ae92734d51bcb82cc35f0a6b5a14928f0036b2c90Jeff Bolzjbolz@nvidia.com2025-08-23T01:31:54-05:00GitHubnoreply@github.com2025-08-23T08:31:54+02:00vulkan: optimize mul_mat_id loading row ids into shared memory (#15427)
838e92734d51bcb82cc35f0a6b5a14928f0036b2c9045363632cbd593537d541e81b600242e0b3d47fcJohannes Gäßlerjohannesg@5d6.de2025-08-22T23:47:01+02:00GitHubnoreply@github.com2025-08-22T23:47:01+02:00test-opt: allow slight inprecision (#15503)
83945363632cbd593537d541e81b600242e0b3d47fc32732f2459a598606055f0403f0e4ec148d06d68Reese Levinereeselevine1@gmail.com2025-08-22T11:28:03-07:00GitHubnoreply@github.com2025-08-22T11:28:03-07:00ggml WebGPU: add support for quantization types (#15440)
84032732f2459a598606055f0403f0e4ec148d06d6892f7f0a53cf6484e16a8084ed90807c35a164809Aldehir Rojashello@alde.dev2025-08-22T11:04:08-05:00GitHubnoreply@github.com2025-08-22T11:04:08-05:00model : gpt-oss add response_format support (#15494)
84192f7f0a53cf6484e16a8084ed90807c35a164809b1ab91821f980f8993423c3f2a82a0a0f60c09d2rmatifrmatif@proton.me2025-08-22T15:33:15+02:00GitHubnoreply@github.com2025-08-22T15:33:15+02:00ggml: add `conv3d` op (#15182)
842b1ab91821f980f8993423c3f2a82a0a0f60c09d29ebebef62fd0adf8685874f154e227ea87b7c6f4Yavor Ivanovyavorgenadiev@gmail.com2025-08-22T14:06:29+03:00GitHubnoreply@github.com2025-08-22T13:06:29+02:00cuda : add Pad Reflect 1D support (#14659)
8439ebebef62fd0adf8685874f154e227ea87b7c6f4ad5c975c2d0297124fad210776ef8eed6b90d578Georgi Gerganovggerganov@gmail.com2025-08-22T12:22:13+03:00GitHubnoreply@github.com2025-08-22T12:22:13+03:00llama : remove KV cache defragmentation logic (#15473)
844ad5c975c2d0297124fad210776ef8eed6b90d5784afb0a746f22abaa545b3ebdb76a400d7da3a713Aaron Teoaaron.teo1@ibm.com2025-08-22T16:11:04+08:00GitHubnoreply@github.com2025-08-22T16:11:04+08:00ggml-cpu: Support Q5_0 and Q5_1 on s390x (#15486)
8454afb0a746f22abaa545b3ebdb76a400d7da3a713e288693669cf9d0a71e2f2b8bd57305f0634025765a10104049+65a@users.noreply.github.com2025-08-22T08:10:14ZGitHubnoreply@github.com2025-08-22T10:10:14+02:00server : Support multimodal completion and embeddings prompts in JSON format (#15108)
846e288693669cf9d0a71e2f2b8bd57305f06340257a0f98dd604d34826eb5ea2560d1e23fe726921dfTarek Dakhrantarek@liquid.ai2025-08-22T09:29:08+02:00GitHubnoreply@github.com2025-08-22T09:29:08+02:00readme : model : mtdm : lfm2 improvements (#15476)
847a0f98dd604d34826eb5ea2560d1e23fe726921df54a241f505d515d625767b993bfd573ecee306b9Chenguang Li757486878@qq.com2025-08-22T14:12:07+08:00GitHubnoreply@github.com2025-08-22T14:12:07+08:00CANN: Optimize RMS_NORM using cache (#15419)
84854a241f505d515d625767b993bfd573ecee306b9cd36b5e5c7fed2a3ac671dd542d579ca40b48b54Diego Devesaslarengh@gmail.com2025-08-21T14:09:32-07:00GitHubnoreply@github.com2025-08-21T23:09:32+02:00sched : fix possible use of wrong ids tensor when offloading moe prompt processing (#15488)
849cd36b5e5c7fed2a3ac671dd542d579ca40b48b543f196be84b1376945737163e35a91e29e3e24d2fGeorgi Gerganovggerganov@gmail.com2025-08-21T19:13:45+03:00GitHubnoreply@github.com2025-08-21T19:13:45+03:00llama : remove deprecated llama_kv_self API (#15472)
8503f196be84b1376945737163e35a91e29e3e24d2f97ae5961a4d9ff9c60f51bac304b97de18e75eafGeorgi Gerganovggerganov@gmail.com2025-08-21T18:44:45+03:00GitHubnoreply@github.com2025-08-21T18:44:45+03:00graph : remove build_attn_with_sinks overload (#15469)
85197ae5961a4d9ff9c60f51bac304b97de18e75eaf20c2dac8c6e05f2ad5295ddef1aebaf2d266090eAclyaclysia@gmail.com2025-08-21T17:01:51+02:00GitHubnoreply@github.com2025-08-21T17:01:51+02:00vulkan : support conv_2d_dw with f16 weights (#15392)
85220c2dac8c6e05f2ad5295ddef1aebaf2d266090e96452a3fa426de83494fb0268a636aa1bfe557feDong Won Kim63934649+ddwkim@users.noreply.github.com2025-08-22T00:00:16+09:00GitHubnoreply@github.com2025-08-21T17:00:16+02:00vulkan: add exp operation (#15456)
85396452a3fa426de83494fb0268a636aa1bfe557fe9ad5e60dba38a6718366b7ac43e7d8e8abdc36c9Jeff Bolzjbolz@nvidia.com2025-08-21T09:55:00-05:00GitHubnoreply@github.com2025-08-21T16:55:00+02:00vulkan: Reuse conversion results in prealloc_y (#15410)
8549ad5e60dba38a6718366b7ac43e7d8e8abdc36c9715a6db02ccb16284837885f2c6fab05d8f7a6eeJie Fu (傅杰)jiefu@tencent.com2025-08-21T22:53:13+08:00GitHubnoreply@github.com2025-08-21T16:53:13+02:00examples : fix some typos in examples/model-conversion/README.md (#15477)
855ad294df03ff2dccd227c3fee653166f3d78b23a4029bb39eb1e7ae0e5df817ce97931abcd5fa52a9Jie Fu (傅杰)jiefu@tencent.com2025-08-21T21:42:34+08:00GitHubnoreply@github.com2025-08-21T15:42:34+02:00examples : install torch-cpu for model conversion tool/example (#15475)
856029bb39eb1e7ae0e5df817ce97931abcd5fa52a930649cab657d87ac46692332a76e1b75d5d22e00Ali Tariqalitariq4589@gmail.com2025-08-21T17:52:16+05:00GitHubnoreply@github.com2025-08-21T14:52:16+02:00ci : enable RVV1.0 native build (#15386)
85730649cab657d87ac46692332a76e1b75d5d22e002758fa10dab0556e6c3f130e664750fd6773dc7cGeorgi Gerganovggerganov@gmail.com2025-08-21T13:42:55+03:00GitHubnoreply@github.com2025-08-21T13:42:55+03:00ci : continue file download with wget (#15471)
8582758fa10dab0556e6c3f130e664750fd6773dc7cb108e429043ee5c9fc8fa4957a0a52c3e490d5c9Daniel Beveniusdaniel.bevenius@gmail.com2025-08-21T12:16:54+02:00GitHubnoreply@github.com2025-08-21T12:16:54+02:00examples : add model conversion tool/example (#15455)
859b108e429043ee5c9fc8fa4957a0a52c3e490d5c9245be739df942861ddc52331b095b40f18e2a3f1Michael Gibamichaelgiba@gmail.com2025-08-21T05:06:46-05:00GitHubnoreply@github.com2025-08-21T12:06:46+02:00ci : fix -Werror=return-type in clip.cpp so ci/run.sh can run without issue (#15221)
860245be739df942861ddc52331b095b40f18e2a3f1b2caf67db1208fd38a0570785c39f7370d906d8aCopilot198982749+Copilot@users.noreply.github.com2025-08-21T11:47:52+02:00GitHubnoreply@github.com2025-08-21T11:47:52+02:00ci : add copilot-instructions.md (#15286)
861b2caf67db1208fd38a0570785c39f7370d906d8a2f3dbffb17ef782edfd50e5a130cec6e8a7e47f8Julien Denize40604584+juliendenize@users.noreply.github.com2025-08-21T11:19:50+02:00GitHubnoreply@github.com2025-08-21T11:19:50+02:00convert : make Mistral community chat templates optional via parameter (#15420)
8622f3dbffb17ef782edfd50e5a130cec6e8a7e47f8945e1f12a6b586ebf82fa4fd7f347225e58174c5Jie Fu (傅杰)jiefu@tencent.com2025-08-21T16:54:34+08:00GitHubnoreply@github.com2025-08-21T11:54:34+03:00common : fix incorrect print of non-ascii characters in the logging (#15466)
863945e1f12a6b586ebf82fa4fd7f347225e58174c51b0db8f6e08951969e2447c2d18bf638effb8f75Xuan-Son Nguyenson@huggingface.co2025-08-21T07:32:26+02:00GitHubnoreply@github.com2025-08-21T08:32:26+03:00ggml : fix condition of im2col on Metal backend (#15460)
8641b0db8f6e08951969e2447c2d18bf638effb8f7529f538ac630d6544406a0702476e36808a6bd1b3stduhpfstephduh@live.fr2025-08-21T07:19:22+02:00GitHubnoreply@github.com2025-08-21T08:19:22+03:00server : fix webui (#15462)
86529f538ac630d6544406a0702476e36808a6bd1b38ad038c0fdc719ced9fbf921a02cbae9ad79287fDaniel Beveniusdaniel.bevenius@gmail.com2025-08-21T06:12:28+02:00GitHubnoreply@github.com2025-08-21T06:12:28+02:00examples : remove references to `make` in examples [no ci] (#15457)
8668ad038c0fdc719ced9fbf921a02cbae9ad79287f5682a3745f2b653dcb855d5766d8edc318fb3336R0CKSTARyeahdongcn@gmail.com2025-08-21T11:06:05+08:00GitHubnoreply@github.com2025-08-21T11:06:05+08:00musa: add GGML_UNUSED_VARS (#15446)
8675682a3745f2b653dcb855d5766d8edc318fb33361bc664a26a1d93a48baf2483a7ff95291ca8bcb8Diego Devesaslarengh@gmail.com2025-08-20T16:35:28-07:00GitHubnoreply@github.com2025-08-21T01:35:28+02:00sched : copy only the used experts when offloading prompt processing (#15346)
8681bc664a26a1d93a48baf2483a7ff95291ca8bcb813aeb7aef284daed4ad07dc14b4b3e2a42b5ea97teoTeoZosa@users.noreply.github.com2025-08-21T07:10:08+09:00GitHubnoreply@github.com2025-08-21T00:10:08+02:00server: fix OpenAI API compatibility for usage statistics in chat streams (#15444)
86913aeb7aef284daed4ad07dc14b4b3e2a42b5ea977a6e91ad26160dd6dfb33d29ac441617422f28e7Johannes Gäßlerjohannesg@5d6.de2025-08-20T23:14:14+02:00GitHubnoreply@github.com2025-08-20T23:14:14+02:00CUDA: refactor FA support/selection code (#15454)
8707a6e91ad26160dd6dfb33d29ac441617422f28e7fec9519802ae1567048abb126cdd5ea160a22d0fJohannes Gäßlerjohannesg@5d6.de2025-08-20T16:58:49+02:00GitHubnoreply@github.com2025-08-20T16:58:49+02:00CUDA: replace GGML_CUDA_F16 with CUDA arch checks (#15433)
871fec9519802ae1567048abb126cdd5ea160a22d0f657b8a77bd01854f99d37a47318fa24f2e7e298fJeff Bolzjbolz@nvidia.com2025-08-20T09:33:14-05:00GitHubnoreply@github.com2025-08-20T16:33:14+02:00vulkan: shorten pipeline name strings (#15431)
872657b8a77bd01854f99d37a47318fa24f2e7e298fec5ab1a36c11dd3efcf4ec8d1ac89a13a8117bc3Daniel Beveniusdaniel.bevenius@gmail.com2025-08-20T14:26:01+02:00GitHubnoreply@github.com2025-08-20T14:26:01+02:00chat: handle gpt-oss return/end token inconsistency (#15421)
873ec5ab1a36c11dd3efcf4ec8d1ac89a13a8117bc31a99c2d948209d9ea5eac8b6dc0a297107244540Jie Fu (傅杰)fujie_email@sina.com2025-08-20T18:33:30+08:00GitHubnoreply@github.com2025-08-20T13:33:30+03:00common : fix context shift help message (#15448)
8741a99c2d948209d9ea5eac8b6dc0a29710724454037f10f955f70e0158d50343d0b9a3f92d194daaexiaobing31871554036+xiaobing318@users.noreply.github.com2025-08-20T18:32:05+08:00GitHubnoreply@github.com2025-08-20T13:32:05+03:00cmake : fix target include directories (#15450)
87537f10f955f70e0158d50343d0b9a3f92d194daae2f37014073f4c6ddc8f241c927db87337c71aa52Daniel Beveniusdaniel.bevenius@gmail.com2025-08-20T12:31:16+02:00GitHubnoreply@github.com2025-08-20T13:31:16+03:00make : remove make in favor of CMake (#15449)
8762f37014073f4c6ddc8f241c927db87337c71aa52a094f381432d92c4bf92d2d6167284316ba73a62Georgi Gerganovggerganov@gmail.com2025-08-20T13:30:46+03:00GitHubnoreply@github.com2025-08-20T13:30:46+03:00lookahead : add sample command to readme (#15447)
877a094f381432d92c4bf92d2d6167284316ba73a62fb22dd07a639e81c7415e30b146f545f1a2f2cafR0CKSTARyeahdongcn@gmail.com2025-08-20T10:17:37+08:00GitHubnoreply@github.com2025-08-20T10:17:37+08:00musa: fix build warnings (#15258)
878fb22dd07a639e81c7415e30b146f545f1a2f2caf9ef6b0b835450ee10cd7be934ad8aef681dc1f43lhezlih@qti.qualcomm.com2025-08-20T02:25:51+08:00GitHubnoreply@github.com2025-08-19T11:25:51-07:00opencl: mark `argsort` unsupported if cols exceed workgroup limit (#15375)
8799ef6b0b835450ee10cd7be934ad8aef681dc1f431e19f5d462b6df490a13103ca555d851f47e5fa5Georgi Gerganovggerganov@gmail.com2025-08-19T19:58:28+03:00GitHubnoreply@github.com2025-08-19T19:58:28+03:00model : add gpt-oss type strings (#15424)
8801e19f5d462b6df490a13103ca555d851f47e5fa5d2fcd91cf96b46f4485ce46b4e3a32bf0df37715Gian-Carlo Pascuttogcp@sjeng.org2025-08-19T18:58:14+02:00GitHubnoreply@github.com2025-08-19T19:58:14+03:00common : Add top-nsigma sampler to help globally (#15428)
881d2fcd91cf96b46f4485ce46b4e3a32bf0df37715a6d3cfe7fa6ea1fb0e1ba8243b731db22ddc0b49Georgi Gerganovggerganov@gmail.com2025-08-19T16:46:37+03:00GitHubnoreply@github.com2025-08-19T16:46:37+03:00server : disable context shift by default (#15416)
882a6d3cfe7fa6ea1fb0e1ba8243b731db22ddc0b4967f09a3a27db443f9870aac87e163dba0d08131eSHUAI YANGshuaiyang047@163.com2025-08-19T21:28:22+08:00GitHubnoreply@github.com2025-08-19T21:28:22+08:00CANN: optimize rope operator (#15335)
88367f09a3a27db443f9870aac87e163dba0d08131e6424594c56f4dbd0573455d89a0d89a0ac093d13R0CKSTARyeahdongcn@gmail.com2025-08-19T18:33:47+08:00GitHubnoreply@github.com2025-08-19T12:33:47+02:00musa: handle __hgt2_mask, available starting from MUSA SDK rc4.3.0 (#15413)
8846424594c56f4dbd0573455d89a0d89a0ac093d13e9288e886970884f288533cd597b3798995b4099Marvin Gießingmarvin.giessing@gmail.com2025-08-19T10:54:31+02:00GitHubnoreply@github.com2025-08-19T11:54:31+03:00ggml-cpu: add mxfp4 VSX intrinsics for Power9+ (ppc64le) hardware (#15385)
885e9288e886970884f288533cd597b3798995b40999d262f4bad0d37838100133537aaf0a83835ed12Xuan-Son Nguyenson@huggingface.co2025-08-19T10:29:36+02:00GitHubnoreply@github.com2025-08-19T10:29:36+02:00chat : clarify the meaning of reasoning_format (#15408)
8869d262f4bad0d37838100133537aaf0a83835ed12f0d3c7405c323784a60f14ddddfbac3f7404d417Georgi Gerganovggerganov@gmail.com2025-08-19T08:45:26+03:00GitHubnoreply@github.com2025-08-19T08:45:26+03:00server : remove swa_full warning (#15399)
887f0d3c7405c323784a60f14ddddfbac3f7404d417f08c4c0d8d0cb6caaf8b7ad316039232b9fa059cGeorgi Gerganovggerganov@gmail.com2025-08-19T08:45:12+03:00GitHubnoreply@github.com2025-08-19T08:45:12+03:00batched-bench : use rand tokens (#15398)
888f08c4c0d8d0cb6caaf8b7ad316039232b9fa059c6d7f1117e3e3285d0c5c11b5ebb0439e27920082Xuan-Son Nguyenson@huggingface.co2025-08-18T22:53:52+02:00GitHubnoreply@github.com2025-08-18T22:53:52+02:00mtmd : clean up clip_n_output_tokens (#15391)
8896d7f1117e3e3285d0c5c11b5ebb0439e2792008260212f1ead2dce9bf1ac69633a7069258ae604d8Georgi Gerganovggerganov@gmail.com2025-08-18T22:02:50+03:00Georgi Gerganovggerganov@gmail.com2025-08-18T22:06:44+03:00codeowners : remove mmv.*
89060212f1ead2dce9bf1ac69633a7069258ae604d8f0c541d315e97b297b3421c52ebde53340ee66b3Georgi Gerganovggerganov@gmail.com2025-08-18T22:02:11+03:00Georgi Gerganovggerganov@gmail.com2025-08-18T22:06:44+03:00sync : ggml
891f0c541d315e97b297b3421c52ebde53340ee66b3baa9255a45105d2d3b4ec432af13b7a6eda3ff35Georgi Gerganovggerganov@gmail.com2025-08-18T20:35:47+03:00Georgi Gerganovggerganov@gmail.com2025-08-18T22:06:44+03:00scripts : update sync scripts
892baa9255a45105d2d3b4ec432af13b7a6eda3ff353007baf201e7ffcda17dbdb0335997fa50a6595bSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-18T19:30:17+02:00GitHubnoreply@github.com2025-08-18T19:30:17+02:00llama : merge conts and reshapes and remove unnecessary cont (#15380)
8933007baf201e7ffcda17dbdb0335997fa50a6595bd1d82416006e7ff41780cb0e9b5f28d30a267497Georgi Gerganovggerganov@gmail.com2025-08-18T18:11:44+03:00GitHubnoreply@github.com2025-08-18T18:11:44+03:00readme : update hot topics (#15397)
894d1d82416006e7ff41780cb0e9b5f28d30a267497618575c5825d7d4f170e686e772178d2aae148aedavidefdavidef1986@gmail.com2025-08-18T16:51:42+02:00GitHubnoreply@github.com2025-08-18T17:51:42+03:00server : fix incoming tasks not process in order (#15395)
895618575c5825d7d4f170e686e772178d2aae148aef44f7931729022c57319a0124931120a169e0da9Dobri Danchev12420863+danchev@users.noreply.github.com2025-08-18T05:50:48-05:00GitHubnoreply@github.com2025-08-18T12:50:48+02:00Fix broken build: require updated pip to support --break-system-packages (#15357)
896f44f7931729022c57319a0124931120a169e0da9ae532eac2c1df1d8edc3d2719145895b966de1bfcompiladegit@compilade.net2025-08-18T03:23:56-04:00GitHubnoreply@github.com2025-08-18T09:23:56+02:00ggml-quants : fix make_qp_quants NANs and IQ1 assertion errors (#15379)
897ae532eac2c1df1d8edc3d2719145895b966de1bfe5155e698645242d4f019267ecc40ea9bad81b09Jeff Bolzjbolz@nvidia.com2025-08-18T00:56:29-05:00GitHubnoreply@github.com2025-08-18T07:56:29+02:00vulkan: disable spirv-opt for bfloat16 shaders (#15352)
898e5155e698645242d4f019267ecc40ea9bad81b0921c17b5befc5f6be5992bc87fc1ba99d388561dfOleksandr Kuvshynov661042+okuvshynov@users.noreply.github.com2025-08-17T18:28:58-04:00GitHubnoreply@github.com2025-08-18T00:28:58+02:00server : export max observed n_past value (#15361)
89921c17b5befc5f6be5992bc87fc1ba99d388561df19f4decae0ead52debe56095ba8d693b4f14e4dfJeff Bolzjbolz@nvidia.com2025-08-17T11:08:57-05:00GitHubnoreply@github.com2025-08-17T18:08:57+02:00vulkan: Use larger workgroups for mul_mat_vec when M is small (#15355)
90019f4decae0ead52debe56095ba8d693b4f14e4df4d196981d4db79e0105b939eaa7ecd40385b721cDong Won Kim63934649+ddwkim@users.noreply.github.com2025-08-17T23:03:09+09:00GitHubnoreply@github.com2025-08-17T16:03:09+02:00vulkan: support sqrt (#15370)
9014d196981d4db79e0105b939eaa7ecd40385b721cb143fbc87af0324aa49e16cd91faf3ba8bb22231Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-17T14:47:42+02:00GitHubnoreply@github.com2025-08-17T14:47:42+02:00convert : force patch_embd weights to F16 or F32 to avoid broken GGUFs (#15367)
902b143fbc87af0324aa49e16cd91faf3ba8bb22231de5627910df74298c998e6bb36ee3217375a5719Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-17T13:30:23+02:00GitHubnoreply@github.com2025-08-17T13:30:23+02:00ci : fix hang in windows-hip build/release (#15365)
903de5627910df74298c998e6bb36ee3217375a571965349f26f2299e06477ec8e85e46243046801358Jeff Bolzjbolz@nvidia.com2025-08-17T03:41:45-05:00GitHubnoreply@github.com2025-08-17T10:41:45+02:00vulkan: Optimize argsort (#15354)
90465349f26f2299e06477ec8e85e462430468013581fe00296f587dfca0957e006d146f5875b61e43dTarek Dakhrant.dakhran@gmail.com2025-08-16T23:33:54+02:00GitHubnoreply@github.com2025-08-16T23:33:54+02:00model : support vision LiquidAI LFM2-VL family (#15347)
9051fe00296f587dfca0957e006d146f5875b61e43dde2192794f4e8e04f2e8167ef2424905145e88fcJeff Bolzjbolz@nvidia.com2025-08-16T11:48:22-05:00GitHubnoreply@github.com2025-08-16T11:48:22-05:00vulkan: fuse adds (#15252)
906de2192794f4e8e04f2e8167ef2424905145e88fc2e2b22ba6607414a5d619ac6d2f034b5b02214e5Jeff Bolzjbolz@nvidia.com2025-08-16T04:18:31-05:00GitHubnoreply@github.com2025-08-16T11:18:31+02:00vulkan: Support mul_mat_id with f32 accumulators (#15337)
9072e2b22ba6607414a5d619ac6d2f034b5b02214e5912ff8c119f01ae029543c7fdf7a84f91a0437a3Jeff Bolzjbolz@nvidia.com2025-08-16T03:58:38-05:00GitHubnoreply@github.com2025-08-16T10:58:38+02:00vulkan: Add missing bounds checking to scalar/coopmat1 mul_mat_id (#15334)
908912ff8c119f01ae029543c7fdf7a84f91a0437a35e6229a8409ac786e62cb133d09f1679a9aec13ermatifkingrealriadh@gmail.com2025-08-16T10:05:55+02:00GitHubnoreply@github.com2025-08-16T01:05:55-07:00OpenCL: add initial FA support (#14987)
9095e6229a8409ac786e62cb133d09f1679a9aec13ee2c1bfff5305c661ac53e9d57cb732ff626a2242Daniel Beveniusdaniel.bevenius@gmail.com2025-08-15T19:50:52+02:00GitHubnoreply@github.com2025-08-15T19:50:52+02:00common : fix double bos, use common_chat_templates for add_bos and add_eos (#15326)
910e2c1bfff5305c661ac53e9d57cb732ff626a22425edf1592fdb9131d01321aeef4241c6a34969e27lhezlih@qti.qualcomm.com2025-08-16T00:52:14+08:00GitHubnoreply@github.com2025-08-15T09:52:14-07:00opencl: add initial mxfp4 support via mv (#15270)
9115edf1592fdb9131d01321aeef4241c6a34969e27db3010bd23980bad5f5d93ec3e6757ec531a413bGeorgi Gerganovggerganov@gmail.com2025-08-15T17:16:36+03:00GitHubnoreply@github.com2025-08-15T16:16:36+02:00vulkan : fix out-of-bounds access in argmax kernel (#15342)
912db3010bd23980bad5f5d93ec3e6757ec531a413bff27f80a74bbe5303acd511a6781a1de6d619b3cGeorgi Gerganovggerganov@gmail.com2025-08-15T16:28:28+03:00GitHubnoreply@github.com2025-08-15T15:28:28+02:00vulkan : fix compile warnings on macos (#15340)
913ff27f80a74bbe5303acd511a6781a1de6d619b3cd3248d9b6557c75d59954c594bb53cf517591e91Aaron Teoaaron.teo1@ibm.com2025-08-15T21:11:22+08:00GitHubnoreply@github.com2025-08-15T21:11:22+08:00ggml: initial IBM zDNN backend (#14975)
914d3248d9b6557c75d59954c594bb53cf517591e917aeee88cfe9c0434eac722b0f7c21404f48758a5Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-15T14:02:39+02:00GitHubnoreply@github.com2025-08-15T14:02:39+02:00ci : fix ios-xcode-build (#15324)
9157aeee88cfe9c0434eac722b0f7c21404f48758a5b07791aa1d4831a08ad54ca19aa206c0c5c0f34aDiego Devesaslarengh@gmail.com2025-08-15T03:27:02-07:00GitHubnoreply@github.com2025-08-15T12:27:02+02:00ci : move ccache action to ggml-org fork (#15328)
916b07791aa1d4831a08ad54ca19aa206c0c5c0f34a4227c9be4268ac844921b90f31595f81236bd317Johannes Gäßlerjohannesg@5d6.de2025-08-15T11:23:17+02:00GitHubnoreply@github.com2025-08-15T11:23:17+02:00test-opt: fix backend support check (#15317)
9174227c9be4268ac844921b90f31595f81236bd317df36bce667bf14f8e538645547754386f9516326Johannes Gäßlerjohannesg@5d6.de2025-08-14T23:21:24+02:00GitHubnoreply@github.com2025-08-14T23:21:24+02:00CUDA: fix negative KV_max values in FA (#15321)
918df36bce667bf14f8e538645547754386f9516326f75b8306472811ecc4e4457d5573a09201f02183Georgi Gerganovggerganov@gmail.com2025-08-14T22:10:51+03:00GitHubnoreply@github.com2025-08-14T22:10:51+03:00eval-callback : stop on first NaN (#15320)
919f75b8306472811ecc4e4457d5573a09201f021837a0de960452f9a57de7f1d167e57b6f3ee5ac1b6Diego Devesaslarengh@gmail.com2025-08-14T10:28:29-07:00GitHubnoreply@github.com2025-08-14T10:28:29-07:00chat : include kwargs in template example (#15309)
9207a0de960452f9a57de7f1d167e57b6f3ee5ac1b6e4e915912cfd2ee15c5a4a0074813232134892f6Daniel Beveniusdaniel.bevenius@gmail.com2025-08-14T17:56:26+02:00GitHubnoreply@github.com2025-08-14T17:56:26+02:00llama : add 18-layer model type for Gemma 3-270m (#15319)
921e4e915912cfd2ee15c5a4a0074813232134892f65ba36f61033d2819be27e2abb70e9a3bd20c0fdesimevogithub@simevo.com2025-08-14T17:45:27+02:00GitHubnoreply@github.com2025-08-14T18:45:27+03:00devops : fix compile bug when the BASE_CUDA_DEV_CONTAINER is based on Ubuntu 24.04 (#15005)
9225ba36f61033d2819be27e2abb70e9a3bd20c0fdeb204a5a234f4cf0b3f449476da639a5510c6d157uvoscarl@uvos.xyz2025-08-14T16:23:56+02:00GitHubnoreply@github.com2025-08-14T16:23:56+02:00HIP: Cleanup hipification header (#15285)
923b204a5a234f4cf0b3f449476da639a5510c6d157646944cfa8961afd914dd6637739b3cda9a72e11Aldehir Rojashello@alde.dev2025-08-14T09:23:11-05:00GitHubnoreply@github.com2025-08-14T17:23:11+03:00gpt-oss: implement harmony parsing (#15181)
924646944cfa8961afd914dd6637739b3cda9a72e111a01899b612ae8f99a174ad076207090e08d4d7bChristian Kastnerckk@kvr.at2025-08-14T16:22:58+02:00GitHubnoreply@github.com2025-08-14T16:22:58+02:00docker : Enable GGML_CPU_ALL_VARIANTS for ARM (#15267)
9251a01899b612ae8f99a174ad076207090e08d4d7b863d341eeb81db104902b14b6f9413daa515e957Georgi Gerganovggerganov@gmail.com2025-08-14T17:16:03+03:00GitHubnoreply@github.com2025-08-14T17:16:03+03:00readme : update hot topics (#15315)
926863d341eeb81db104902b14b6f9413daa515e957d32e03f4495d3efa1c5126f53b449f1d429c5664Jeff Bolzjbolz@nvidia.com2025-08-14T08:38:10-05:00GitHubnoreply@github.com2025-08-14T08:38:10-05:00vulkan: perf_logger improvements (#15246)
927d32e03f4495d3efa1c5126f53b449f1d429c56643973163bff40f7f5161b0f08a0011729e2b0406aGeorgi Gerganovggerganov@gmail.com2025-08-14T14:59:50+03:00GitHubnoreply@github.com2025-08-14T14:59:50+03:00server : add SWA checkpoints (#15293)
9283973163bff40f7f5161b0f08a0011729e2b0406a5ade3000bd6040bf6878eafd6c77168552f73c47Georgi Gerganovggerganov@gmail.com2025-08-14T14:19:23+03:00Georgi Gerganovggerganov@gmail.com2025-08-14T14:59:27+03:00sync : ggml
9295ade3000bd6040bf6878eafd6c77168552f73c478b2483730f90d6f23e2b188a54a210498bba937fJason Nijason.ni.py@gmail.com2025-08-14T19:17:51+08:00Georgi Gerganovggerganov@gmail.com2025-08-14T14:59:27+03:00ggml: fix ggml_conv_1d_dw bug (ggml/1323)
9308b2483730f90d6f23e2b188a54a210498bba937f810b9fc8b99dd55517a3e94c6dee29748d482559Georgi Gerganovggerganov@gmail.com2025-08-14T13:41:03+03:00Georgi Gerganovggerganov@gmail.com2025-08-14T14:59:27+03:00tests : remove unused includes (ggml/0)
931810b9fc8b99dd55517a3e94c6dee29748d4825594ebd0c125b24a0d7a78b0ffc1d9567530ed8f0c4kallewoofkarljohan-alm@garage.co.jp2025-08-14T20:03:30+09:00GitHubnoreply@github.com2025-08-14T14:03:30+03:00perplexity : provide a helpful hint for has_cpl case in split_equal error. (#15304)
9324ebd0c125b24a0d7a78b0ffc1d9567530ed8f0c45cdb27e0917479d2d742cea7beee089574bb09faSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-14T12:22:07+02:00GitHubnoreply@github.com2025-08-14T13:22:07+03:00cuda : fix GGML_CUDA_GRAPHS=OFF (#15300)
9335cdb27e0917479d2d742cea7beee089574bb09fa3ea913f1ce9567289aedd866a569dbab8fb8e419Jonathan Graehl99024+graehl@users.noreply.github.com2025-08-14T03:03:57-07:00GitHubnoreply@github.com2025-08-14T12:03:57+02:00finetune: SGD optimizer, more CLI args (#13873)
9343ea913f1ce9567289aedd866a569dbab8fb8e41929c8fbe4e05fd23c44950d0958299e25fbeabc5ckallewoofkarljohan-alm@garage.co.jp2025-08-14T15:16:32+09:00GitHubnoreply@github.com2025-08-14T09:16:32+03:00perplexity: give more information about constraints on failure (#15303)
93529c8fbe4e05fd23c44950d0958299e25fbeabc5c1adc9812bd33dc85489bf093528d61c22917d54fuvoscarl@uvos.xyz2025-08-13T20:44:30+02:00GitHubnoreply@github.com2025-08-13T20:44:30+02:00HIP: bump requirement to rocm 6.1 (#15296)
9361adc9812bd33dc85489bf093528d61c22917d54fb3e16665e14032d1276f9d0263acef8321b6f518Bas Nijholtbasnijholt@gmail.com2025-08-13T11:21:31-07:00GitHubnoreply@github.com2025-08-13T11:21:31-07:00fix(nix): remove non-functional llama-cpp cachix cache from flake.nix (#15295)
937b3e16665e14032d1276f9d0263acef8321b6f518c24f4e26883a91219af5acfaf1471ca7ef582683Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-13T15:43:00+02:00GitHubnoreply@github.com2025-08-13T15:43:00+02:00server : enable -td and -tbd parameters (#15172)
938c24f4e26883a91219af5acfaf1471ca7ef582683d8914fc47e8a69b28c670325cb1c8ce33e3a2960Judd4046440+foldl@users.noreply.github.com2025-08-13T18:45:15+08:00GitHubnoreply@github.com2025-08-13T13:45:15+03:00ggml : update `ggml_rope_multi` (#12665)
939d8914fc47e8a69b28c670325cb1c8ce33e3a2960e885445bc1719d2e289a9b2e11714e0f994e68beCopilot198982749+Copilot@users.noreply.github.com2025-08-13T12:44:40+02:00GitHubnoreply@github.com2025-08-13T12:44:40+02:00 common : add --override-tensor-draft, --cpu-moe-draft and --n-cpu-moe-draft parameters (#15191)
940e885445bc1719d2e289a9b2e11714e0f994e68be648ebcdb739abfb5a9be14f4c5c0fd19ff268ef0Aldehir Rojashello@alde.dev2025-08-13T05:28:21-05:00GitHubnoreply@github.com2025-08-13T12:28:21+02:00server : filter out harmony thought messages (#15278)
941648ebcdb739abfb5a9be14f4c5c0fd19ff268ef007aa869a91837d95fcb5612c65a188763ac38647Ali Tariqalitariq4589@gmail.com2025-08-13T15:14:44+05:00GitHubnoreply@github.com2025-08-13T13:14:44+03:00ci : Added CI with RISC-V RVV1.0 Hardware (#14439)
94207aa869a91837d95fcb5612c65a188763ac3864700f35d509e5367bf19ccaf4b4adddc38db4811c6Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-13T11:30:45+02:00GitHubnoreply@github.com2025-08-13T11:30:45+02:00ci : add more python requirements to copilot-setup-steps (#15289)
94300f35d509e5367bf19ccaf4b4adddc38db4811c66028bf74351d35a06bd98498624f8c2f029f7d1aGeorgi Gerganovggerganov@gmail.com2025-08-13T11:09:39+03:00GitHubnoreply@github.com2025-08-13T11:09:39+03:00ggml : repack block_iq4_nlx8 (#14904)
9446028bf74351d35a06bd98498624f8c2f029f7d1abc5182272c373267352bc689e5fca276934bea2dOliver Simonsosimons@nvidia.com2025-08-13T10:04:46+02:00GitHubnoreply@github.com2025-08-13T10:04:46+02:00CUDA: Optimize `reduce_rows_f32` kernel, leading up to 25x perf improvement on kernel-level and 10% perf increase for Gemma3n (#15132)
945bc5182272c373267352bc689e5fca276934bea2de71d48e3265027351e44a8e198f933c98f242c2eSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-13T09:07:13+02:00GitHubnoreply@github.com2025-08-13T09:07:13+02:00ci : add copilot-setup-steps.yml (#15214)
946e71d48e3265027351e44a8e198f933c98f242c2eb0493156fa8622694f21f42460a84da3eded0bc0Tak-RSsnosk.t@gmail.com2025-08-13T14:54:30+09:00GitHubnoreply@github.com2025-08-13T08:54:30+03:00ggml-rpc: chunk send()/recv() to avoid EINVAL for very large tensors over RPC (macOS & others) (#15188)
9473a617cb18121a85eed17ac5585788810c5d9c1e0c42293848dd9f03eba9a6c1b85a600b398f06541Yunzhe Jiayunzhe@calculet.tech2025-08-13T10:58:59+08:00Yunzhe Jiayunzhe@calculet.tech2025-08-13T10:58:59+08:00add calrt_decode
948b0493156fa8622694f21f42460a84da3eded0bc0f4586ee5986d6f965becb37876d6f3666478a961uvoscarl@uvos.xyz2025-08-12T22:15:12+02:00GitHubnoreply@github.com2025-08-12T22:15:12+02:00HIP: disable sync warp shuffel operators from clr amd_warp_sync_functions.h (#15273)
949f4586ee5986d6f965becb37876d6f3666478a96160a76588106d22ccacd6b1a0d15d8545861d0a0dRomain Biessyromain.biessy@codeplay.com2025-08-12T13:58:22+02:00GitHubnoreply@github.com2025-08-12T13:58:22+02:00sycl: Fix and disable more configurations of mul_mat (#15151)
95060a76588106d22ccacd6b1a0d15d8545861d0a0defe3a90996ca6e67ca90f337fc03ad551082c408rmatifkingrealriadh@gmail.com2025-08-12T11:42:41+02:00GitHubnoreply@github.com2025-08-12T02:42:41-07:00opencl: allow mixed f16/f32 `add` (#15140)
951efe3a90996ca6e67ca90f337fc03ad551082c408bbd57b7eafb3b32e3f7f3a2175bce0b35abc7de8Aman Guptaamangupta052@gmail.com2025-08-12T17:21:45+08:00GitHubnoreply@github.com2025-08-12T17:21:45+08:00CUDA cmake: add `-lineinfo` for easier debug (#15260)
952bbd57b7eafb3b32e3f7f3a2175bce0b35abc7de825ff6f7659f6a5c47d6a73eada5813f0495331f0Chenguang Li757486878@qq.com2025-08-12T16:12:13+08:00GitHubnoreply@github.com2025-08-12T16:12:13+08:00CANN: GGML_OP_CPY optimization (#15070)
95325ff6f7659f6a5c47d6a73eada5813f0495331f0be48528b068111304e4a0bb82c028558b5705f05R0CKSTARyeahdongcn@gmail.com2025-08-12T10:02:51+08:00GitHubnoreply@github.com2025-08-12T10:02:51+08:00musa: fix failures in test-backend-ops for mul_mat_id op (#15236)
954be48528b068111304e4a0bb82c028558b5705f05cf9e5648a7ae02f7728fefcbcfd2979d83c96e8fhipuddinghuafengchun@gmail.com2025-08-11T22:50:31+08:00GitHubnoreply@github.com2025-08-11T22:50:31+08:00CANN: Add broadcast for softmax and FA (#15208)
955cf9e5648a7ae02f7728fefcbcfd2979d83c96e8ffba5c0d680b555cbac563fef57b33bc5c9621e08rainred107027757+gryffindor-rr@users.noreply.github.com2025-08-11T22:12:12+08:00GitHubnoreply@github.com2025-08-11T16:12:12+02:00mtmd : Fix MinicpmV model converter and clip to avoid using hardcode. (#14750)
956fba5c0d680b555cbac563fef57b33bc5c9621e0853d0a1265826f40c5dbc01d06aeab9e14fcbd69bXuan-Son Nguyenson@huggingface.co2025-08-11T15:31:35+02:00GitHubnoreply@github.com2025-08-11T15:31:35+02:00chat : hotfix gpt-oss jinja raising an exception (#15243)
95753d0a1265826f40c5dbc01d06aeab9e14fcbd69b27093afe78912494073eb043fec93a007e49653cXuan-Son Nguyenson@huggingface.co2025-08-11T14:48:41+02:00GitHubnoreply@github.com2025-08-11T14:48:41+02:00server : allow specifying reasoning_format in HTTP request (#15238)
95827093afe78912494073eb043fec93a007e49653c228f724d9ce6c56e8cec75bfdabab4dd013def7fZagajm.zagajewska@gmail.com2025-08-11T14:27:54+02:00GitHubnoreply@github.com2025-08-11T15:27:54+03:00readme : update infra list (#15234)
959228f724d9ce6c56e8cec75bfdabab4dd013def7fcd3069dfcbeee8e0e96cffb93b0ebb9e595e273aGeorgi Gerganovggerganov@gmail.com2025-08-11T13:58:24+03:00GitHubnoreply@github.com2025-08-11T13:58:24+03:00kv-cache : fix seq_rm with seq_id == -1 (#15226)
960cd3069dfcbeee8e0e96cffb93b0ebb9e595e273a50e81bdf5db563ab57a9a722b08f96fa8a76c927Daniel Beveniusdaniel.bevenius@gmail.com2025-08-11T11:21:19+02:00GitHubnoreply@github.com2025-08-11T11:21:19+02:00kv-cache : log (debug) all streams in find_slot (#15176)
96150e81bdf5db563ab57a9a722b08f96fa8a76c9271ebbaddff2a44b0599df659175d3274bd5bbeb81Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-11T11:15:44+02:00GitHubnoreply@github.com2025-08-11T11:15:44+02:00convert : fix merge conflicts (#15229)
9621ebbaddff2a44b0599df659175d3274bd5bbeb81a3a7874272e5a060079658eb5cca4617b7f99062Daniel Beveniusdaniel.bevenius@gmail.com2025-08-11T10:21:24+02:00GitHubnoreply@github.com2025-08-11T11:21:24+03:00perplexity : update comments/error msg to use decode [no ci] (#15227)
963a3a7874272e5a060079658eb5cca4617b7f99062002cb1bb3345967fbe0fa7766c2d94c2da31ef45Julien Denize40604584+juliendenize@users.noreply.github.com2025-08-11T10:07:49+02:00GitHubnoreply@github.com2025-08-11T10:07:49+02:00convert : improve Mistral models integration (#14737)
964002cb1bb3345967fbe0fa7766c2d94c2da31ef4579c1160b073b8148a404f3dd2584be1606dccc66Charles Xucharles.xu@arm.com2025-08-11T09:59:26+02:00GitHubnoreply@github.com2025-08-11T09:59:26+02:00kleidiai: fix unsigned overflow bug (#15150)
96579c1160b073b8148a404f3dd2584be1606dccc6634c9d765bf173c551398f1e7fa4595019bc53babDavid Zhao90013954+Your-Cheese@users.noreply.github.com2025-08-09T13:29:43-05:00GitHubnoreply@github.com2025-08-09T20:29:43+02:00cuda: refactored ssm_scan and use CUB (#13291)
96634c9d765bf173c551398f1e7fa4595019bc53babe54d41befcc1575f4c898c5ff4ef43970cead75fAman Guptaamangupta052@gmail.com2025-08-09T20:00:24+08:00GitHubnoreply@github.com2025-08-09T20:00:24+08:00CUDA: add attention sinks for tile and wmma (#15178)
967e54d41befcc1575f4c898c5ff4ef43970cead75f4850b52aedceeb70bb4fe49f2d7cd1df6ee98682compiladegit@compilade.net2025-08-08T17:48:26-04:00GitHubnoreply@github.com2025-08-08T17:48:26-04:00gguf-py : add Numpy MXFP4 de/quantization support (#15111)
9684850b52aedceeb70bb4fe49f2d7cd1df6ee98682cd6983d56d2cce94ecb86bb114ae8379a609073cJohannes Gäßlerjohannesg@5d6.de2025-08-08T23:04:36+02:00GitHubnoreply@github.com2025-08-08T23:04:36+02:00server-bench: external OAI servers, sqlite (#15179)
969cd6983d56d2cce94ecb86bb114ae8379a609073c6c7e9a54406dbba5e53754a8f70a285414717b06AN Longaisk@users.noreply.github.com2025-08-08T21:37:22+09:00GitHubnoreply@github.com2025-08-08T14:37:22+02:00ggml : fix field name when new ggml_backend (#14944)
9706c7e9a54406dbba5e53754a8f70a285414717b061425f587a82bc303469b5c32759a2746ba4e1e20Olivier Chafikolivier.chafik@gmail.com2025-08-08T10:45:18+01:00GitHubnoreply@github.com2025-08-08T10:45:18+01:00vendor: sync minja (#15161)
9711425f587a82bc303469b5c32759a2746ba4e1e20aaa3d07ae749b781d6135eaff23c7fa8a4ab404aJohannes Gäßlerjohannesg@5d6.de2025-08-08T08:19:58+02:00GitHubnoreply@github.com2025-08-08T08:19:58+02:00CUDA: attention sinks for mma FlashAttention (#15157)
972aaa3d07ae749b781d6135eaff23c7fa8a4ab404a50aa9389014bba2dd12234132aa6b8ca3601a17flhezlih@qti.qualcomm.com2025-08-08T13:47:03+09:00GitHubnoreply@github.com2025-08-07T21:47:03-07:00opencl: support sink in `soft_max` (attn sinks) (#15152)
97350aa9389014bba2dd12234132aa6b8ca3601a17fc4f53563df4575196ea13f5ed669ea8ea659a6beXuan-Son Nguyenson@huggingface.co2025-08-07T23:26:03+02:00GitHubnoreply@github.com2025-08-07T23:26:03+02:00convert : support non-mxfp4 HF model (#15153)
974c4f53563df4575196ea13f5ed669ea8ea659a6bea0552c8beef74e843bb085c8ef0c63f9ed7a2b27Jeff Bolzjbolz@nvidia.com2025-08-07T15:44:20-05:00GitHubnoreply@github.com2025-08-07T22:44:20+02:00vulkan: support fattn sinks (#15126)
975a0552c8beef74e843bb085c8ef0c63f9ed7a2b2799acbc9921b119aa7ed929eb5780a66a8f06e6d9Jeff Bolzjbolz@nvidia.com2025-08-07T15:07:11-05:00GitHubnoreply@github.com2025-08-07T22:07:11+02:00vulkan: Add env var to disable host visible vidmem (#15109)
97699acbc9921b119aa7ed929eb5780a66a8f06e6d97ad67ba9fe2b909e271dd31b99c5fce3aba35899RunningLeonmnsheng@yeah.net2025-08-08T00:20:40+08:00GitHubnoreply@github.com2025-08-07T18:20:40+02:00llama : Support intern-s1 (#14875)
9777ad67ba9fe2b909e271dd31b99c5fce3aba358999a96389544a08fd829fccda28142ce2066017fdeuvoscarl@uvos.xyz2025-08-07T16:44:14+02:00GitHubnoreply@github.com2025-08-07T16:44:14+02:00HIP: add cmake option to enable compiler output of kernel resource usage metrics (#15103)
9789a96389544a08fd829fccda28142ce2066017fde1d72c841888b9450916bdd5a9b3274da380f5b36Christian Kastnerckk@kvr.at2025-08-07T13:45:41+02:00GitHubnoreply@github.com2025-08-07T13:45:41+02:00ggml: Skip backend library linking code when GGML_BACKEND_DL=ON (#15094)
9791d72c841888b9450916bdd5a9b3274da380f5b3620638e4f16fcc21f836c7556e83bbf532bb5a0f0Johannes Gäßlerjohannesg@5d6.de2025-08-07T10:53:21+02:00GitHubnoreply@github.com2025-08-07T10:53:21+02:00CUDA: GEMM for FP32/FP16/BF16 and ne11 <= 16 (#15131)
98020638e4f16fcc21f836c7556e83bbf532bb5a0f036d3f00e142696f708ab297b9f8f1c825594712dJohannes Gäßlerjohannesg@5d6.de2025-08-07T08:50:30+02:00GitHubnoreply@github.com2025-08-07T08:50:30+02:00scripts: fix crash when --tool is not set (#15133)
98136d3f00e142696f708ab297b9f8f1c825594712d5fd160bbd9d70b94b5b11b0001fd7f477005e4a0Daniel Beveniusdaniel.bevenius@gmail.com2025-08-07T05:31:48+02:00GitHubnoreply@github.com2025-08-07T05:31:48+02:00requirements : fix PyTorch uint64 compatibility (#15134)
9825fd160bbd9d70b94b5b11b0001fd7f477005e4a0756cfea82608911bbfcbf45164b8fdaddbafaa31Reese Levinereeselevine1@gmail.com2025-08-06T15:14:40-07:00GitHubnoreply@github.com2025-08-06T15:14:40-07:00ggml: Add basic SET_ROWS support in WebGPU (#15137)
983756cfea82608911bbfcbf45164b8fdaddbafaa31e725a1a982ca870404a9c4935df52466327bbd02rmatifkingrealriadh@gmail.com2025-08-06T23:17:51+02:00GitHubnoreply@github.com2025-08-06T14:17:51-07:00fix profiling crash (#15072)
984e725a1a982ca870404a9c4935df52466327bbd023db4da56a5a00c4b47bbb3b18f85fb4473662adblhezlih@qti.qualcomm.com2025-08-07T04:12:17+09:00GitHubnoreply@github.com2025-08-06T12:12:17-07:00opencl: add `swiglu_oai` and `add_id` (#15121)
9853db4da56a5a00c4b47bbb3b18f85fb4473662adb476aa3fd5779b32d06cd84338f777da82341195cSachin Desaismdesai@gmail.com2025-08-06T11:27:30-07:00GitHubnoreply@github.com2025-08-06T20:27:30+02:00chat : support Granite model reasoning and tool call (#14864)
986476aa3fd5779b32d06cd84338f777da82341195c0d8831543cdc368fb248bae6f1b4aa5516684edcJuk Armstrong69222624+jukofyork@users.noreply.github.com2025-08-06T17:28:48+01:00GitHubnoreply@github.com2025-08-06T17:28:48+01:00Fixed name `-override-tensors` to `-override-tensor` (#15129)
9870d8831543cdc368fb248bae6f1b4aa5516684edc65c797c4fad4d9966695ac4b4a1560be44109267Diego Devesaslarengh@gmail.com2025-08-06T05:37:35-07:00GitHubnoreply@github.com2025-08-06T14:37:35+02:00ggml : fix fallback to CPU for ununsupported ops (#15118)
98865c797c4fad4d9966695ac4b4a1560be4410926725726898e855ec6dffba227f2233a63c57184036Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-06T13:26:49+02:00GitHubnoreply@github.com2025-08-06T13:26:49+02:00chat : fix yandex chat template (#15116)
98925726898e855ec6dffba227f2233a63c571840362241453252147bb7362a286977ee9f9a92130062stevenkuangstevenkuang@tencent.com2025-08-06T17:48:30+08:00GitHubnoreply@github.com2025-08-06T11:48:30+02:00chat : fix hunyuan auto-detection (#15114)
990c42293848dd9f03eba9a6c1b85a600b398f065411cbf442f50a154aac0bf35d7b8b3436a2cdfd6f5Yunzhe Jiayunzhe@calculet.tech2025-08-06T15:46:06+08:00Yunzhe Jiayunzhe@calculet.tech2025-08-06T15:46:06+08:00sync with calrt API
9911cbf442f50a154aac0bf35d7b8b3436a2cdfd6f5ba67d6c03fed3193987a4ef13050e6e2a4451df4Yunzhe Jiayunzhe@calculet.tech2025-07-24T15:50:19+08:00Yunzhe Jiayunzhe@calculet.tech2025-08-06T15:45:10+08:001.sync with calrt prefill,decode,kv_update op 2.add hw pattern untile function
9922241453252147bb7362a286977ee9f9a921300629515c6131aecaccc955fdedcfe16c3e030aaefcbChenguang Li757486878@qq.com2025-08-06T14:12:42+08:00GitHubnoreply@github.com2025-08-06T14:12:42+08:00CANN: add support for ACL Graph (#15065)
9939515c6131aecaccc955fdedcfe16c3e030aaefcbfd1234cb468935ea087d6929b2487926c3afff4bReese Levinereeselevine1@gmail.com2025-08-05T16:26:38-07:00GitHubnoreply@github.com2025-08-05T16:26:38-07:00ggml: WebGPU disable SET_ROWS for now (#15078)
994fd1234cb468935ea087d6929b2487926c3afff4bf324a3b715d5c1081c110ce459f8a8486fb1ee89Georgi Gerganovggerganov@gmail.com2025-08-05T22:10:36+03:00GitHubnoreply@github.com2025-08-05T22:10:36+03:00llama : add gpt-oss (#15091)
995f324a3b715d5c1081c110ce459f8a8486fb1ee89be426425817bc3e6a2d91dae476dba6fa85894beSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-05T20:43:36+02:00GitHubnoreply@github.com2025-08-05T20:43:36+02:00chat : only remove double bos/eos if added (#15086)
996be426425817bc3e6a2d91dae476dba6fa85894be3306ceabf02e3df66666e5851800e843c7ca207eGeorgi Gerganovggerganov@gmail.com2025-08-05T20:19:33+03:00GitHubnoreply@github.com2025-08-05T20:19:33+03:00readme : update hot topics (#15097)
9973306ceabf02e3df66666e5851800e843c7ca207ec81de6e107ef51ef76aadcb8a6f008711c462517Romain Biessyromain.biessy@codeplay.com2025-08-05T18:39:55+02:00GitHubnoreply@github.com2025-08-05T18:39:55+02:00sycl: fix mul_mat selection (#15092)
998c81de6e107ef51ef76aadcb8a6f008711c46251722f060c9c4b5ef49a83a20eda25fcb792419580bJuk Armstrong69222624+jukofyork@users.noreply.github.com2025-08-05T13:56:44+01:00GitHubnoreply@github.com2025-08-05T13:56:44+01:00Fix `glm4moe` bug (#15088)
99922f060c9c4b5ef49a83a20eda25fcb792419580bee3a9fcf88fe5b5e1213711e05861b83cd4fdfe6Alex Wudindinw@users.noreply.github.com2025-08-05T19:56:44+08:00GitHubnoreply@github.com2025-08-05T13:56:44+02:00webui: fix markdown table (#15081)
1000ee3a9fcf88fe5b5e1213711e05861b83cd4fdfe6ec428b02c347767f24c78111309e3f30d2ada289compiladegit@compilade.net2025-08-05T05:27:45-04:00GitHubnoreply@github.com2025-08-05T11:27:45+02:00context : fix index overflow on huge outputs (#15080)
1001ec428b02c347767f24c78111309e3f30d2ada28919f68fa5a4c3bf796de52a6db9008e77d29f423aDiego Devesaslarengh@gmail.com2025-08-04T16:05:36-07:00GitHubnoreply@github.com2025-08-05T01:05:36+02:00llama : add --n-cpu-moe option (#15077)
100219f68fa5a4c3bf796de52a6db9008e77d29f423a41613437ffee0dbccad684fc744788bc504ec213compiladegit@compilade.net2025-08-04T17:26:52-04:00GitHubnoreply@github.com2025-08-04T23:26:52+02:00imatrix : warn when GGUF imatrix is saved without .gguf suffix (#15076)
100341613437ffee0dbccad684fc744788bc504ec213e5bebe5251cee2678e8531aa1598ca21b3c6ce1dChristian Kastnerckk@kvr.at2025-08-04T21:29:14+02:00GitHubnoreply@github.com2025-08-04T21:29:14+02:00cmake: Add GGML_BACKEND_DIR option (#15074)
1004e5bebe5251cee2678e8531aa1598ca21b3c6ce1def0144c087b33e5b8da42d529ac71aaf05cb49dfSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-04T21:01:48+02:00GitHubnoreply@github.com2025-08-04T21:01:48+02:00gguf-py : add --chat-template-file to gguf_new_metadata (#15075)
1005ef0144c087b33e5b8da42d529ac71aaf05cb49df2721257e3e2c4c944ac8a08221113ee7cb503f1bSamsammcj@users.noreply.github.com2025-08-05T04:29:25+10:00GitHubnoreply@github.com2025-08-04T20:29:25+02:00model: support GLM 4.5 family of models (#14939)
10062721257e3e2c4c944ac8a08221113ee7cb503f1b587d0118f50b7e8f4bafbcdd218aefd9da0272e1Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-04T18:11:02+02:00GitHubnoreply@github.com2025-08-04T18:11:02+02:00quantize : fix confusing error message if ftype is invalid (#15071)
1007587d0118f50b7e8f4bafbcdd218aefd9da0272e15aa1105da24a8dd1661cea3db0582c9b2c2f54d3Reese Levinereeselevine1@gmail.com2025-08-04T08:52:43-07:00GitHubnoreply@github.com2025-08-04T08:52:43-07:00ggml: WebGPU backend host improvements and style fixing (#14978)
10085aa1105da24a8dd1661cea3db0582c9b2c2f54d3d31192b4ee1441bbbecd3cbf9e02633368bdc4f5Jeff Bolzjbolz@nvidia.com2025-08-04T00:09:19-05:00GitHubnoreply@github.com2025-08-04T07:09:19+02:00vulkan: fix build when using glslang that does not support coopmat2 (#15062)
1009d31192b4ee1441bbbecd3cbf9e02633368bdc4f50a2f5496bef9e54e5f42d6c2c3ad9eb7b379aed0compiladegit@compilade.net2025-08-03T16:00:05-04:00GitHubnoreply@github.com2025-08-03T22:00:05+02:00imatrix : use GGUF by default (#14842)
10100a2f5496bef9e54e5f42d6c2c3ad9eb7b379aed011a3811164ef2d75393c6b0a632f4c608e3e3dd2compiladegit@compilade.net2025-08-03T15:49:13-04:00GitHubnoreply@github.com2025-08-03T21:49:13+02:00imatrix : fix 3d activation handling for hybrid and recurrent models (#14994)
101111a3811164ef2d75393c6b0a632f4c608e3e3dd297366dc6abdd0bdc74260bd3c42bd06f0feb7428compiladegit@compilade.net2025-08-03T15:43:07-04:00GitHubnoreply@github.com2025-08-03T21:43:07+02:00memory : handle kv_unified for hybrid models (#15050)
101297366dc6abdd0bdc74260bd3c42bd06f0feb742883bc2f288c0e08e676d9beca9c4669197e920593Csaba Kecskemeticsaba.kecskemeti@gmail.com2025-08-03T12:38:18-07:00GitHubnoreply@github.com2025-08-03T21:38:18+02:00vocab : JetBrains Mellum pre-tokenizer (#15045)
101383bc2f288c0e08e676d9beca9c4669197e9205936c7a441161080551ce8a52ba32563b6295067192Gabriel Larson55459720+gabriellarson@users.noreply.github.com2025-08-03T09:56:25-05:00GitHubnoreply@github.com2025-08-03T16:56:25+02:00model : add text-only support for Kimi-VL (and find special tokens in text_config) (#15051)
10146c7a441161080551ce8a52ba32563b62950671925c0eb5ef544aeefd81c303e03208f768e158d93cJeff Bolzjbolz@nvidia.com2025-08-03T07:23:57-05:00GitHubnoreply@github.com2025-08-03T14:23:57+02:00vulkan: Use coopmat2 for conv2d (#14982)
10155c0eb5ef544aeefd81c303e03208f768e158d93c03d46982180c2fb624bd2a233e46426ab22be5d1lhezlih@qti.qualcomm.com2025-08-02T10:51:18-07:00GitHubnoreply@github.com2025-08-02T19:51:18+02:00opencl: fix adreno compiler detection logic (#15029)
101603d46982180c2fb624bd2a233e46426ab22be5d13303c19b1691088275ee864a823697177c94a15dJohannes Gäßlerjohannesg@5d6.de2025-08-02T16:37:08+02:00GitHubnoreply@github.com2025-08-02T16:37:08+02:00CUDA: use mma FA kernel for gqa > 4 on RTX 4000 (#15035)
10173303c19b1691088275ee864a823697177c94a15d4fdea540bda4648f98b85e8ee9dc66db4bfb5945leejetleejet714@gmail.com2025-08-02T22:15:36+08:00GitHubnoreply@github.com2025-08-02T17:15:36+03:00cuda: make im2col a little faster (#15025)
10184fdea540bda4648f98b85e8ee9dc66db4bfb5945a4569c41fd2253c89ef52fc2378687bdbf42f61aDaniel Beveniusdaniel.bevenius@gmail.com2025-08-02T16:14:57+02:00GitHubnoreply@github.com2025-08-02T17:14:57+03:00kv-cache : skip alignment of n_stream in kv-cache log msg [no ci] (#15040)
1019a4569c41fd2253c89ef52fc2378687bdbf42f61a15e92fd33791e60a4ddb5970b47242a855c27117Georgi Gerganovggerganov@gmail.com2025-08-02T17:14:21+03:00GitHubnoreply@github.com2025-08-02T17:14:21+03:00llama : enable LLAMA_SET_ROWS=1 by default (#14959)
102015e92fd33791e60a4ddb5970b47242a855c271172bf3fbf0b54f97aef2b388b76d222789e1c170f1Georgi Gerganovggerganov@gmail.com2025-08-02T17:13:05+03:00GitHubnoreply@github.com2025-08-02T17:13:05+03:00cuda, sycl : fix batched gemm when ne02 == 1 && ne03 > 1 (#15038)
10212bf3fbf0b54f97aef2b388b76d222789e1c170f1711d5e6fe66eb6cd7a10d71cec4567321848be08Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-08-02T14:39:01+02:00GitHubnoreply@github.com2025-08-02T14:39:01+02:00ci : check that pre-tokenizer hashes are up-to-date (#15032)
1022711d5e6fe66eb6cd7a10d71cec4567321848be08f738989dcb9ccbe468c945553eafbeef7b869675Douglas Hanleythesecretaryofwar@gmail.com2025-08-02T05:51:02-05:00GitHubnoreply@github.com2025-08-02T12:51:02+02:00convert : fix Qwen3-Embedding pre-tokenizer hash (#15030)
1023f738989dcb9ccbe468c945553eafbeef7b8696754cb208c93c1c938591a5b40354e2a6f9b94489bcJhen-Jie Hongiainst0409@gmail.com2025-08-02T18:04:48+08:00GitHubnoreply@github.com2025-08-02T18:04:48+08:00chat : fix multiple tool_calls on hermes-2-pro (#14962)
10244cb208c93c1c938591a5b40354e2a6f9b94489bc3025b621d12a6931ff5e9775d4f644719980ad91Jeff Bolzjbolz@nvidia.com2025-08-02T04:21:37-05:00GitHubnoreply@github.com2025-08-02T11:21:37+02:00vulkan: coopmat2 mul_mat optimizations (#14934)
10253025b621d12a6931ff5e9775d4f644719980ad91ec0b18802c91badd3ff1388ffd09ee163251bd72R0CKSTARyeahdongcn@gmail.com2025-08-02T17:20:40+08:00GitHubnoreply@github.com2025-08-02T17:20:40+08:00llama-bench: rename DB table name from test to llama_bench (#15003)
1026ec0b18802c91badd3ff1388ffd09ee163251bd72339bd0268c498c89529cd0e90c44883c211e3745Jeff Bolzjbolz@nvidia.com2025-08-02T03:48:30-05:00GitHubnoreply@github.com2025-08-02T10:48:30+02:00vulkan: Support ne[3]>1 in noncontig matrix-vector multiply (#15015)
1027339bd0268c498c89529cd0e90c44883c211e3745f906275537d14c8fc7c6976d944233771fd6672cDouglas Hanleythesecretaryofwar@gmail.com2025-08-02T03:44:50-05:00GitHubnoreply@github.com2025-08-02T10:44:50+02:00model : support Qwen3-Embedding (#15023)
1028f906275537d14c8fc7c6976d944233771fd6672ca9f7541ec25c4c8547daf5ff48700ad2836e2b7dJohannes Gäßlerjohannesg@5d6.de2025-08-02T10:12:41+02:00GitHubnoreply@github.com2025-08-02T10:12:41+02:00server: enable token array inputs for OAI API (#15001)
1029a9f7541ec25c4c8547daf5ff48700ad2836e2b7d9c35706b98ea271858acef4194f526a71b24cdc9Jeff Bolzjbolz@nvidia.com2025-08-02T02:57:04-05:00GitHubnoreply@github.com2025-08-02T09:57:04+02:00vulkan: optimizations for direct convolution (#14933)
10309c35706b98ea271858acef4194f526a71b24cdc9c76b420e4ce06f7b7cdfbb0b85d02c90e5cc5a3aJohannes Gäßlerjohannesg@5d6.de2025-08-01T20:47:32+02:00GitHubnoreply@github.com2025-08-01T20:47:32+02:00CUDA: fix MMQ nwarps for AMD with warp_size==32 (#15014)
1031c76b420e4ce06f7b7cdfbb0b85d02c90e5cc5a3a0f5ccd6fd1a1f709010312933db0316867cc30b6l-austenfeld53152202+l-austenfeld@users.noreply.github.com2025-08-01T16:59:06+02:00GitHubnoreply@github.com2025-08-01T16:59:06+02:00vendor : update vendored copy of google/minja (#15011)
10320f5ccd6fd1a1f709010312933db0316867cc30b61c872f71fb8a25589efa3ee9b6bf8b517cb8caa4stevenkuangstevenkuang@tencent.com2025-08-01T21:31:12+08:00GitHubnoreply@github.com2025-08-01T15:31:12+02:00model : add hunyuan dense (#14878)
10331c872f71fb8a25589efa3ee9b6bf8b517cb8caa4baad94885df512bb24ab01e2b22d1998fce4d00elhezquic_lih@quicinc.com2025-08-01T04:15:44-07:00GitHubnoreply@github.com2025-08-01T13:15:44+02:00opencl: add f16 for `add`, `sub`, `mul`, `div` (#14984)
1034baad94885df512bb24ab01e2b22d1998fce4d00eba42794c9ead96ad52311ba1b23eefcbf3d6f63dSrihari-mcw96763064+Srihari-mcw@users.noreply.github.com2025-08-01T11:50:33+05:30GitHubnoreply@github.com2025-08-01T09:20:33+03:00ggml : Q2k interleaving implementation - x86/x64 SIMD (#14373)
1035ba42794c9ead96ad52311ba1b23eefcbf3d6f63d2860d479b456e1caa026b40b829d5b13c42a8ed7Georgi Gerganovggerganov@gmail.com2025-08-01T06:38:12+03:00GitHubnoreply@github.com2025-08-01T06:38:12+03:00graph : fix equal_seq() check (#14986)
10362860d479b456e1caa026b40b829d5b13c42a8ed7484b2091ce5017901483b5204c07878f171d1441diannao55k@outlook.com2025-08-01T10:02:34+08:00GitHubnoreply@github.com2025-08-01T10:02:34+08:00docker : add cann build pipline (#14591)
1037484b2091ce5017901483b5204c07878f171d1441daf2dd788066b8b239cb7f68210e090c2124c199R0CKSTARyeahdongcn@gmail.com2025-08-01T08:47:27+08:00GitHubnoreply@github.com2025-08-01T08:47:27+08:00compare-commits.sh: support both llama-bench and test-backend-ops (#14392)
1038daf2dd788066b8b239cb7f68210e090c2124c199a06ed5feaec9f935fbf662035b2673167bc88460Ed Addario29247825+EAddario@users.noreply.github.com2025-07-31T20:32:18+01:00GitHubnoreply@github.com2025-07-31T21:32:18+02:00quantize : skip tensor override when in fallback mode (#14995)
1039a06ed5feaec9f935fbf662035b2673167bc88460784524053d986255e383dae45235e4f76c6792a7Diego Devesaslarengh@gmail.com2025-07-31T11:15:41-07:00GitHubnoreply@github.com2025-07-31T20:15:41+02:00llama : add simple option to enable CPU for MoE weights (--cpu-moe) (#14992)
1040784524053d986255e383dae45235e4f76c6792a7d6818d06a6237631523bc0f45d42e79482667948Aman Guptaamangupta052@gmail.com2025-08-01T01:22:58+08:00GitHubnoreply@github.com2025-08-01T01:22:58+08:00Fix params bug in diffusion example (#14993)
1041d6818d06a6237631523bc0f45d42e79482667948e08a98826bcce70a3377592c51d0efed99eafe07Diego Devesaslarengh@gmail.com2025-07-31T09:11:34-07:00GitHubnoreply@github.com2025-07-31T18:11:34+02:00llama : allow other bufts when overriding to CPU, add --no-repack option (#14990)
1042e08a98826bcce70a3377592c51d0efed99eafe07952a47f455fbd92e2659b98b9b6317a2dafeb532Ruben Ortlampicard12@live.de2025-07-31T17:46:54+02:00GitHubnoreply@github.com2025-07-31T17:46:54+02:00Vulkan: Fix minor debug mode issues (#14899)
1043952a47f455fbd92e2659b98b9b6317a2dafeb53236e5fe7bcd6640f4ff974d4c5cb04a13b5b29bcetc-mb157115220+tc-mb@users.noreply.github.com2025-07-31T23:22:17+08:00GitHubnoreply@github.com2025-07-31T17:22:17+02:00mtmd : support MiniCPM-V 4.0 (#14983)
104436e5fe7bcd6640f4ff974d4c5cb04a13b5b29bce94933c8c2eeaa9a7983e3f6c08af76bd86724094Csaba Kecskemeticsaba.kecskemeti@gmail.com2025-07-31T07:59:49-07:00GitHubnoreply@github.com2025-07-31T10:59:49-04:00MODEL_TENSOR.SSM_DT_NORM has defined twice (#14991)
104594933c8c2eeaa9a7983e3f6c08af76bd86724094c1dacaa99b4ead6edbac928cd2da59436573f6b0g2mt166577174+g2mt@users.noreply.github.com2025-07-31T05:25:23-07:00GitHubnoreply@github.com2025-07-31T14:25:23+02:00server : implement universal assisted decoding (#12635)
1046c1dacaa99b4ead6edbac928cd2da59436573f6b0a9f77a8be348deeb11fb3d54d412bf583003c90dDongliang Wei121270393+wdl339@users.noreply.github.com2025-07-31T20:12:20+08:00GitHubnoreply@github.com2025-07-31T14:12:20+02:00llama : merge build_moe_ffn_from_probs function into build_moe_ffn (#14968)
1047a9f77a8be348deeb11fb3d54d412bf583003c90d8a4a85627702b569d7d2810f2de06a4321656e9dLukas Straublukasstraub2@web.de2025-07-31T14:08:23+02:00GitHubnoreply@github.com2025-07-31T14:08:23+02:00server : add openai-style logit_bias support (#14946)
10488a4a85627702b569d7d2810f2de06a4321656e9d11490b36723d511d75fb601995c79b5c363ba3a2Aman Guptaamangupta052@gmail.com2025-07-31T19:49:09+08:00GitHubnoreply@github.com2025-07-31T19:49:09+08:00Add LLaDA 8b Diffusion model (#14771)
104911490b36723d511d75fb601995c79b5c363ba3a266625a59a54d0a7504eda4c4e83abfcd83ba1cf8hipuddinghuafengchun@gmail.com2025-07-31T19:47:20+08:00GitHubnoreply@github.com2025-07-31T19:47:20+08:00CANN: Improve loading efficiency after converting weights to NZ format. (#14985)
105066625a59a54d0a7504eda4c4e83abfcd83ba1cf86e6725459a892b49602b596339de4916c7c7965acompiladegit@compilade.net2025-07-31T01:02:46-04:00GitHubnoreply@github.com2025-07-31T08:02:46+03:00graph : reduce splits for recurrent and hybrid models (#14825)
10516e6725459a892b49602b596339de4916c7c7965ae9192bec564780bd4313ad6524d20a0ab92797dblhezlih@qti.qualcomm.com2025-07-30T14:56:55-07:00GitHubnoreply@github.com2025-07-30T14:56:55-07:00opencl: add `mul_mat_f32_f32_l4_lm` and `mul_mat_f16_f32_l4_lm` (#14809)
1052e9192bec564780bd4313ad6524d20a0ab92797db41e78c567e9a8c652e405f4f909deb598deecd31Ed Addario29247825+EAddario@users.noreply.github.com2025-07-30T20:11:56+01:00GitHubnoreply@github.com2025-07-30T21:11:56+02:00quantize : fix using combined imatrix GGUFs (multiple datasets) (#14973)
105341e78c567e9a8c652e405f4f909deb598deecd31ad4a700117d1746799d2d6599e526e2c3a7938d2Daniel Beveniusdaniel.bevenius@gmail.com2025-07-30T18:07:11+02:00GitHubnoreply@github.com2025-07-30T18:07:11+02:00server : add support for `embd_normalize` parameter (#14964)
1054ad4a700117d1746799d2d6599e526e2c3a7938d2e32a4ec60ee472acf3fe82b5966976fd6965ac6buvoscarl@uvos.xyz2025-07-30T17:38:06+02:00GitHubnoreply@github.com2025-07-30T17:38:06+02:00HIP: enable mfma mmq on gfx908 and gfx90a for select datatypes and shapes (#14949)
1055e32a4ec60ee472acf3fe82b5966976fd6965ac6be228de94496636681b36690247d96f97d5f76c0dGeorgi Gerganovggerganov@gmail.com2025-07-30T16:03:13+03:00Georgi Gerganovggerganov@gmail.com2025-07-30T17:33:11+03:00sync : ggml
1056e228de94496636681b36690247d96f97d5f76c0d73a8e5ca0372f4dcaf1aed4e42261723da0914aaKai Pastordg0yt@darc.de2025-07-30T14:53:16+02:00Georgi Gerganovggerganov@gmail.com2025-07-30T17:33:11+03:00cmake : Fix BLAS link interface (ggml/1316)
105773a8e5ca0372f4dcaf1aed4e42261723da0914aa92b8810ec7aa6d778bc287cc918443cf67b962e2Kai Pastordg0yt@darc.de2025-07-30T14:52:26+02:00Georgi Gerganovggerganov@gmail.com2025-07-30T17:33:11+03:00vulkan : fix 32-bit builds (ggml/1313)
105892b8810ec7aa6d778bc287cc918443cf67b962e200131d6eaf4df029e1ec84de868c2c5957503007Johannes Gäßlerjohannesg@5d6.de2025-07-30T15:46:13+02:00GitHubnoreply@github.com2025-07-30T15:46:13+02:00CUDA: skip masked KV slices for all FA kernels (#14924)
105900131d6eaf4df029e1ec84de868c2c59575030071e15bfd42c3938506e0da5939bf7f42780965f01Georgi Gerganovggerganov@gmail.com2025-07-30T15:12:02+03:00GitHubnoreply@github.com2025-07-30T15:12:02+03:00tests : update for LLAMA_SET_ROWS=1 (#14961)
10601e15bfd42c3938506e0da5939bf7f42780965f01a118d80233d3bf92569c051346fd2638f87bf202Georgi Gerganovggerganov@gmail.com2025-07-30T13:52:11+03:00GitHubnoreply@github.com2025-07-30T13:52:11+03:00graph : fix stack-use-after-return (#14960)
1061a118d80233d3bf92569c051346fd2638f87bf20261550f8231dc0aa478e2f537c5009ede6878ce22Douglas Hanleythesecretaryofwar@gmail.com2025-07-30T00:25:05-05:00GitHubnoreply@github.com2025-07-30T08:25:05+03:00embeddings: fix extraction of CLS pooling results (#14927)
106261550f8231dc0aa478e2f537c5009ede6878ce22aa79524c51fb014f8df17069d31d7c44b9ea6cb8Xinpeng Dou15529241576@163.com2025-07-30T08:39:24+08:00GitHubnoreply@github.com2025-07-30T08:39:24+08:00CANN: update ops docs (#14935)
1063aa79524c51fb014f8df17069d31d7c44b9ea6cb8b77d11179d7efe68ce5c913006f2ef9ee17cc8f7uvoscarl@uvos.xyz2025-07-29T20:23:04+02:00GitHubnoreply@github.com2025-07-29T20:23:04+02:00HIP: remove the use of __HIP_PLATFORM_AMD__, explicitly support only AMD targets (#14945)
1064b77d11179d7efe68ce5c913006f2ef9ee17cc8f7c7aa1364fd59b2ac06fd9e0a719253d968472dc3uvoscarl@uvos.xyz2025-07-29T17:44:30+02:00GitHubnoreply@github.com2025-07-29T17:44:30+02:00HIP: add GGML_HIP_MMQ_MFMA option to allow disableing the MFMA path. (#14930)
1065c7aa1364fd59b2ac06fd9e0a719253d968472dc31a67fcc30677e96dda76bb1b290788e7d8852b51uvoscarl@uvos.xyz2025-07-29T17:43:43+02:00GitHubnoreply@github.com2025-07-29T17:43:43+02:00HIP: Ignore unsupported unroll transformation in fattn-vec (#14931)
10661a67fcc30677e96dda76bb1b290788e7d8852b51204f2cf168dc01ca7b200b1510e0ff585ca9a92ekallewoofkarljohan-alm@garage.co.jp2025-07-30T00:05:38+09:00GitHubnoreply@github.com2025-07-29T17:05:38+02:00common : avoid logging partial messages (which can contain broken UTF-8 sequences) (#14937)
1067204f2cf168dc01ca7b200b1510e0ff585ca9a92e138b288b594eeb19d58e0ab0d74eae32009c9c20hipuddinghuafengchun@gmail.com2025-07-29T22:36:43+08:00GitHubnoreply@github.com2025-07-29T22:36:43+08:00CANN: Add ggml_set_rows (#14943)
1068138b288b594eeb19d58e0ab0d74eae32009c9c20bbd0f917797e9d524680f1b30d34a46eb06d7651Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-29T14:22:03+02:00GitHubnoreply@github.com2025-07-29T14:22:03+02:00cuda : add softcap fusion (#14907)
1069bbd0f917797e9d524680f1b30d34a46eb06d76510a5036bee9cfb946870689db4400e9e0d17844c9Johannes Gäßlerjohannesg@5d6.de2025-07-29T10:40:50+02:00GitHubnoreply@github.com2025-07-29T10:40:50+02:00server-bench: make seed choice configurable (#14929)
10700a5036bee9cfb946870689db4400e9e0d17844c98ad7b3e65b5834e5574c2f5640056c9047b5d93bAman Guptaamangupta052@gmail.com2025-07-29T14:45:18+08:00GitHubnoreply@github.com2025-07-29T14:45:18+08:00CUDA: add roll (#14919)
10718ad7b3e65b5834e5574c2f5640056c9047b5d93bbda62193b2a6bebbf515c3c389303094a44458c1lhezlih@qti.qualcomm.com2025-07-28T09:50:17-07:00GitHubnoreply@github.com2025-07-28T18:50:17+02:00opencl : add ops docs (#14910)
1072bda62193b2a6bebbf515c3c389303094a44458c1c556418b600ad5792440942079d93e393595688bLeonard Mosescutlemo@users.noreply.github.com2025-07-28T09:04:27-07:00GitHubnoreply@github.com2025-07-28T18:04:27+02:00test-backend-ops : extend test case filtering (#14865)
1073c556418b600ad5792440942079d93e393595688bdb16e2831c0f344f041af3d067db81c42b16eb22Radoslav Gerganovrgerganov@gmail.com2025-07-28T18:59:04+03:00GitHubnoreply@github.com2025-07-28T18:59:04+03:00llama-bench : use local GPUs along with RPC servers (#14917)
1074db16e2831c0f344f041af3d067db81c42b16eb22cd1fce6d4f9c191f1c7429cc96f61281c3b63ffcxctanxc-tan@outlook.com2025-07-28T23:40:24+08:00GitHubnoreply@github.com2025-07-28T17:40:24+02:00ggml-cpu : deduplicate scalar implementations (#14897)
1075cd1fce6d4f9c191f1c7429cc96f61281c3b63ffc00fa15fedc79263fa0285e6a3bbb0cfb3e3878a2Akarshan Biswasakarshan@menlo.ai2025-07-28T20:32:15+05:30GitHubnoreply@github.com2025-07-28T20:32:15+05:30SYCL: Add set_rows support for quantized types (#14883)
107600fa15fedc79263fa0285e6a3bbb0cfb3e3878a2946b1f685909c8c9c044f145bce819c02f327eaaXuan-Son Nguyenson@huggingface.co2025-07-28T15:01:48+02:00GitHubnoreply@github.com2025-07-28T15:01:48+02:00mtmd : add support for Voxtral (#14862)
1077946b1f685909c8c9c044f145bce819c02f327eaa6c6e397affc4fac717e718364fb4b635cec6433aJohannes Gäßlerjohannesg@5d6.de2025-07-28T14:30:22+02:00GitHubnoreply@github.com2025-07-28T14:30:22+02:00CUDA: fix pointer incrementation in FA (#14916)
10786c6e397affc4fac717e718364fb4b635cec6433aafc0e8969896ada62238da07b98731e5a4b12ba4Dongliang Wei121270393+wdl339@users.noreply.github.com2025-07-28T19:47:00+08:00GitHubnoreply@github.com2025-07-28T13:47:00+02:00model : add support for SmallThinker series (#14898)
1079afc0e8969896ada62238da07b98731e5a4b12ba4a5771c9eea801f573dc7375416e4c3306543563dAlberto Cabrera Pérezalberto.cabrera@codeplay.com2025-07-28T11:05:53+01:00GitHubnoreply@github.com2025-07-28T11:05:53+01:00sycl: refactor quantization to q8_1 (#14815)
1080a5771c9eea801f573dc7375416e4c3306543563dc35f9eaf095e0db3aa9b77b56846bab5d3bf5661Georgi Gerganovggerganov@gmail.com2025-07-28T11:01:03+03:00GitHubnoreply@github.com2025-07-28T10:01:03+02:00ops : update BLAS (#14914)
1081c35f9eaf095e0db3aa9b77b56846bab5d3bf56611f45f2890ef7f365ba0a45e08a8d1f46b8bc6b9eGeorgi Gerganovggerganov@gmail.com2025-07-28T08:22:56+03:00GitHubnoreply@github.com2025-07-28T08:22:56+03:00ops : update Metal (#14912)
10821f45f2890ef7f365ba0a45e08a8d1f46b8bc6b9e613c5095c33de9b7bbf1097d7c32510f51d58b01Georgi Gerganovggerganov@gmail.com2025-07-28T08:14:20+03:00Georgi Gerganovggerganov@gmail.com2025-07-28T08:15:01+03:00sync : ggml
1083613c5095c33de9b7bbf1097d7c32510f51d58b017f97599581fcf0c37432dd3b1f503b91bed97695Kai Pastordg0yt@darc.de2025-07-24T19:58:02+02:00Georgi Gerganovggerganov@gmail.com2025-07-28T08:15:01+03:00cmake : Indent ggml-config.cmake (ggml/1310)
10847f97599581fcf0c37432dd3b1f503b91bed97695bf78f5439ee8e82e367674043303ebf8e92b4805Ed Addario29247825+EAddario@users.noreply.github.com2025-07-27T22:31:11+01:00GitHubnoreply@github.com2025-07-27T23:31:11+02:00quantize : update README.md (#14905)
1085bf78f5439ee8e82e367674043303ebf8e92b4805bbfc84927481d1e59d8af6939b93b76850f0ab53Ruben Ortlampicard12@live.de2025-07-27T15:33:08+02:00GitHubnoreply@github.com2025-07-27T15:33:08+02:00vulkan: add ops docs (#14900)
1086bbfc84927481d1e59d8af6939b93b76850f0ab53ca0ef2dddb022cb1337d775cd05cd27d7808aff4Akarshan Biswasakarshan@menlo.ai2025-07-27T17:52:58+05:30GitHubnoreply@github.com2025-07-27T17:52:58+05:30SYCL: add ops doc (#14901)
1087ca0ef2dddb022cb1337d775cd05cd27d7808aff489d1029559bd2968f76db854f9f113d73e34527cDaniel Beveniusdaniel.bevenius@gmail.com2025-07-27T12:10:51+02:00GitHubnoreply@github.com2025-07-27T12:10:51+02:00llama : clarify comment about pp and tg graphs [no ci] (#14895)
108889d1029559bd2968f76db854f9f113d73e34527cf1a4e72de5950ab7136aeadddd675caf30dd6b3fErik ScholzGreen-Sky@users.noreply.github.com2025-07-27T12:04:33+02:00GitHubnoreply@github.com2025-07-27T12:04:33+02:00vulkan : add fp16 support for the conv_2d kernel (#14872)
1089f1a4e72de5950ab7136aeadddd675caf30dd6b3f4762ad7316dcdec20016ab5985fb46a27902204dJeff Bolzjbolz@nvidia.com2025-07-27T04:05:34-05:00GitHubnoreply@github.com2025-07-27T11:05:34+02:00vulkan: skip empty set_rows to avoid invalid API usage (#14860)
10904762ad7316dcdec20016ab5985fb46a27902204d1dc9614e0673e794d2e2bf88ba04f7d57b63a57bGabriel Larson55459720+gabriellarson@users.noreply.github.com2025-07-27T03:18:37-05:00GitHubnoreply@github.com2025-07-27T11:18:37+03:00model : make rope_yarn_log_mul optional for deepseek2 (#14896)
10911dc9614e0673e794d2e2bf88ba04f7d57b63a57b446595b9b3a113d9ba10506922c3a156cca9d477Shunta Saitoshunta.saito@gmail.com2025-07-27T16:38:44+09:00GitHubnoreply@github.com2025-07-27T09:38:44+02:00llama : fix kq_scale for the attention layers of PLaMo2 (#14892)
1092446595b9b3a113d9ba10506922c3a156cca9d47766906cd82a4a1fd10151707cee3f66cb61fc4055Aman Guptaamangupta052@gmail.com2025-07-27T09:36:43+08:00GitHubnoreply@github.com2025-07-27T09:36:43+08:00Docs: add instructions for adding backends (#14889)
109366906cd82a4a1fd10151707cee3f66cb61fc405511dd5a44eb180e1d69fac24d3852b5222d66fb7fdeepsek166548550+deepsek@users.noreply.github.com2025-07-26T18:28:14-04:00GitHubnoreply@github.com2025-07-27T00:28:14+02:00HIP: Enable Matrix cores for MMQ Kernels, Enable stream-K for CDNA 3 (#14624)
109411dd5a44eb180e1d69fac24d3852b5222d66fb7f9b8f3c6c776d77045ee4f7f13fdf7863dfe59e5bhipuddinghuafengchun@gmail.com2025-07-26T17:56:18+08:00GitHubnoreply@github.com2025-07-26T17:56:18+08:00CANN: Implement GLU ops (#14884)
10959b8f3c6c776d77045ee4f7f13fdf7863dfe59e5bc7f3169cd523140a288095f2d79befb20a0b73f4R0CKSTARyeahdongcn@gmail.com2025-07-26T10:36:02+08:00GitHubnoreply@github.com2025-07-26T10:36:02+08:00musa: fix build warnings (unused variable) (#14869)
1096c7f3169cd523140a288095f2d79befb20a0b73f4793c0d7f46384001738c337d7afa46b45ae32745Aaron Teoaaron.teo1@ibm.com2025-07-26T01:09:03+08:00GitHubnoreply@github.com2025-07-25T19:09:03+02:00ggml-cpu : disable GGML_NNPA by default due to instability (#14880)
1097793c0d7f46384001738c337d7afa46b45ae32745ce111d39d666f4a3b6a561ad020f6feb8cc67790Gabe Goodhartghart@us.ibm.com2025-07-25T10:47:39-06:00GitHubnoreply@github.com2025-07-25T10:47:39-06:00metal: SSM_SCAN performance (#14743)
1098ce111d39d666f4a3b6a561ad020f6feb8cc67790e7fecba93416c8aaf343932b5e36dd5e208a815elhezlih@qti.qualcomm.com2025-07-25T08:12:13-07:00GitHubnoreply@github.com2025-07-25T17:12:13+02:00opencl: add fused `rms_norm_mul` (#14841)
1099e7fecba93416c8aaf343932b5e36dd5e208a815ee2b7621e7c265a6739225125cf9c534f471b3472wooksongwook16.song@samsung.com2025-07-25T23:25:05+09:00GitHubnoreply@github.com2025-07-25T16:25:05+02:00docs : update HOWTO‑add‑model.md for ModelBase and new model classes (#14874)
1100e2b7621e7c265a6739225125cf9c534f471b3472c1dbea752a630169c104118fd0c82f3ffcb19c91Oliver Simonsosimons@nvidia.com2025-07-25T13:29:57+02:00GitHubnoreply@github.com2025-07-25T14:29:57+03:00ggml : remove invalid portPos specifiers from dot files (#14838)
1101c1dbea752a630169c104118fd0c82f3ffcb19c91749e0d27f0247337869f4698f59dd7fafba94326Georgi Gerganovggerganov@gmail.com2025-07-25T14:28:06+03:00GitHubnoreply@github.com2025-07-25T14:28:06+03:00context : restore preemptive sched reset when LLAMA_SET_ROWS=0 (#14870)
1102749e0d27f0247337869f4698f59dd7fafba9432664bf1c3744053cf7def10aeed21ff48883ee755bkiwi122582483+kiwi142857@users.noreply.github.com2025-07-25T19:08:04+08:00GitHubnoreply@github.com2025-07-25T13:08:04+02:00mtmd : fix 32-bit narrowing issue in export-lora and mtmd clip (#14503)
110364bf1c3744053cf7def10aeed21ff48883ee755bc12bbde37258c6d053322d1cedd4a9672109ae58Chris Rohlfchris.rohlf@gmail.com2025-07-25T06:17:02-04:00GitHubnoreply@github.com2025-07-25T12:17:02+02:00rpc : check for null buffers in get/set/copy tensor endpoints (#14868)
1104c12bbde37258c6d053322d1cedd4a9672109ae583f4fc97f1d745f1d5d3c853949503136d419e6deDiego Devesaslarengh@gmail.com2025-07-25T01:07:26-07:00GitHubnoreply@github.com2025-07-25T11:07:26+03:00sched : fix multiple evaluations of the same graph with pipeline parallelism (#14855)
11053f4fc97f1d745f1d5d3c853949503136d419e6de2df255da3cea108de0ae9b302ffdd31674b6d88dR0CKSTARyeahdongcn@gmail.com2025-07-25T03:05:37+08:00GitHubnoreply@github.com2025-07-24T20:05:37+01:00musa: upgrade musa sdk to rc4.2.0 (#14498)
11062df255da3cea108de0ae9b302ffdd31674b6d88d60f816a79dd74007158745530e71738aa6caa67eGeorgi Gerganovggerganov@gmail.com2025-07-24T18:30:33+03:00Georgi Gerganovggerganov@gmail.com2025-07-24T20:27:23+03:00sync : ggml
110760f816a79dd74007158745530e71738aa6caa67e5592f278b6ec6aa4a1793e89e8c61838a12ebc9dKai Pastordg0yt@darc.de2025-07-22T20:13:21+02:00Georgi Gerganovggerganov@gmail.com2025-07-24T20:27:23+03:00cmake : fix usage issues (ggml/1257)
11085592f278b6ec6aa4a1793e89e8c61838a12ebc9de4868d16d24dec55e61bcaadaca28feed8f98b13Daniel Beveniusdaniel.bevenius@gmail.com2025-07-21T15:53:12+02:00Georgi Gerganovggerganov@gmail.com2025-07-24T20:27:23+03:00ggml-cpu : remove stdlib include from repack.cpp (ggml/1276)
1109e4868d16d24dec55e61bcaadaca28feed8f98b13820de57d4faa427a3d0bfb14e48057247fae036eGeorgi Gerganovggerganov@gmail.com2025-07-24T16:31:48+03:00GitHubnoreply@github.com2025-07-24T16:31:48+03:00context : perform output reorder lazily upon access after sync (#14853)
1110820de57d4faa427a3d0bfb14e48057247fae036ecb4a63aad6650c2b536a7578403935388cb2920eXuan-Son Nguyenson@huggingface.co2025-07-24T13:59:56+02:00GitHubnoreply@github.com2025-07-24T13:59:56+02:00chat : fix kimi-k2 chat template (#14852)
1111cb4a63aad6650c2b536a7578403935388cb2920e86f5623d904cfd392fdeb14a143097b4074660f6Alberto Cabrera Pérezalberto.cabrera@codeplay.com2025-07-24T11:09:57+01:00GitHubnoreply@github.com2025-07-24T11:09:57+01:00sycl: fixed semantics of block offset calculation (#14814)
111286f5623d904cfd392fdeb14a143097b4074660f639cffdf18855e0d2beba62572542251d87421e73yummy57988893+jk3456a@users.noreply.github.com2025-07-24T17:50:51+08:00GitHubnoreply@github.com2025-07-24T11:50:51+02:00llama : fix MiniCPM inference after Granite Four changes (#14850)
111339cffdf18855e0d2beba62572542251d87421e73065908cb09adff8b2a5f1173f80050d5d9a6790bPouyaPooyaGhahramanian@Gmail.com2025-07-24T12:26:44+03:00GitHubnoreply@github.com2025-07-24T11:26:44+02:00docs: add libcurl-dev install hint for Linux distros (#14801)
1114065908cb09adff8b2a5f1173f80050d5d9a6790b4ec6291a2407404de52239c1f9ca66c07e7fb28bGeorgi Gerganovggerganov@gmail.com2025-07-24T10:24:05+03:00GitHubnoreply@github.com2025-07-24T10:24:05+03:00metal : fix fusion across different encoders (#14849)
11154ec6291a2407404de52239c1f9ca66c07e7fb28ba12363bbf0ca11f787dee9756861043136edc0dfDonghyeon Jeong54725479+djeong20@users.noreply.github.com2025-07-24T13:50:41+09:00GitHubnoreply@github.com2025-07-24T12:50:41+08:00sycl: fix undefined variable in work group size check (#14843)
1116a12363bbf0ca11f787dee9756861043136edc0dfa86f52b2859dae4db5a7a0bbc0f1ad9de6b43ec6jacekpoplawski67507230+jacekpoplawski@users.noreply.github.com2025-07-23T23:23:57+02:00GitHubnoreply@github.com2025-07-23T23:23:57+02:00convert : text-only support for GLM-4.1V-9B-Thinking (#14823)
1117a86f52b2859dae4db5a7a0bbc0f1ad9de6b43ec6b284197df426fb189cdcfe56a43c863a788ac756Johannes Gäßlerjohannesg@5d6.de2025-07-23T21:43:25+02:00GitHubnoreply@github.com2025-07-23T21:43:25+02:00CUDA: fix overflow in FA, tune performance (#14840)
1118b284197df426fb189cdcfe56a43c863a788ac756221c0e0c5841a814e95b9bfd549de9d6ae00ac6eJohannes Gäßlerjohannesg@5d6.de2025-07-23T18:22:30+02:00GitHubnoreply@github.com2025-07-23T18:22:30+02:00CUDA: fix compilation with GGML_CUDA_F16 (#14837)
1119221c0e0c5841a814e95b9bfd549de9d6ae00ac6e07a19e27a26f76d34be62da53807f93131fb3cabSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-23T14:27:54+02:00GitHubnoreply@github.com2025-07-23T14:27:54+02:00ci : correct label refactor->refactoring (#14832)
112007a19e27a26f76d34be62da53807f93131fb3cab18f3b5ff9e5eda4e7d04bceff8ffdccb0a696ed8Johannes Gäßlerjohannesg@5d6.de2025-07-23T12:35:53+02:00Georgi Gerganovggerganov@gmail.com2025-07-23T14:08:09+03:00CUDA: fix quantized KV cache + multiple sequences (#14822)
112118f3b5ff9e5eda4e7d04bceff8ffdccb0a696ed87233358d29df3dfbf80693f2de7e5865401ad724Georgi Gerganovggerganov@gmail.com2025-07-18T13:36:27+03:00Georgi Gerganovggerganov@gmail.com2025-07-23T14:08:09+03:00tests : add non-cont K,V FA tests
11227233358d29df3dfbf80693f2de7e5865401ad7246c88b3bb2509d980e6a64c50fdf8dd304929f770l3utterflygc.pthzfoldr@gmail.com2025-07-23T16:16:41+08:00GitHubnoreply@github.com2025-07-23T11:16:41+03:00memory : handle saving/loading null layers in recurrent memory (#14675)
11236c88b3bb2509d980e6a64c50fdf8dd304929f77014c28dfc50a2e3915e697122cd3c5343224009belixing-star104126818+lixing-star@users.noreply.github.com2025-07-23T14:39:51+08:00GitHubnoreply@github.com2025-07-23T09:39:51+03:00ggml: fix loongarch quantize_row_q8_1 error (#14827)
112414c28dfc50a2e3915e697122cd3c5343224009be8c988fa41db4d9dbc05abfd666c04def1259c645chen fan350211548@qq.com2025-07-23T11:58:00+08:00GitHubnoreply@github.com2025-07-23T11:58:00+08:00CANN: weight format to NZ for Ascend310P3 (#14407)
11258c988fa41db4d9dbc05abfd666c04def1259c645acd6cb1c41676f6bbb25c2a76fa5abeb1719301eAman Guptaamangupta052@gmail.com2025-07-23T09:25:42+08:00GitHubnoreply@github.com2025-07-23T09:25:42+08:00CUDA: add fused rms norm (#14800)
1126acd6cb1c41676f6bbb25c2a76fa5abeb1719301e84712b60439453fb393c2ca753cee682a4ad41f5Csaba Kecskemeticsaba.kecskemeti@gmail.com2025-07-22T09:29:43-07:00GitHubnoreply@github.com2025-07-22T19:29:43+03:00ggml : model card yaml tab->2xspace (#14819)
112784712b60439453fb393c2ca753cee682a4ad41f5d4d1522b20809a350ffb094db20f40f17d3ab80fJeff Bolzjbolz@nvidia.com2025-07-22T10:35:21-05:00GitHubnoreply@github.com2025-07-22T17:35:21+02:00vulkan: fix rms_norm_mul to handle broadcasting dim0 (#14817)
1128d4d1522b20809a350ffb094db20f40f17d3ab80fd1aa0cc5d13ec3fa553de5d298f9166679e58479Molly Sophiamollysophia379@gmail.com2025-07-22T23:01:29+08:00GitHubnoreply@github.com2025-07-22T23:01:29+08:00llama : add model type detection for rwkv7 7B&14B (#14816)
1129d1aa0cc5d13ec3fa553de5d298f9166679e58479c8ade30036139e32108fee53d8b7164dbfda4beeEd Addario29247825+EAddario@users.noreply.github.com2025-07-22T13:33:37+01:00GitHubnoreply@github.com2025-07-22T14:33:37+02:00imatrix: add option to display importance score statistics for a given imatrix file (#12718)
1130c8ade30036139e32108fee53d8b7164dbfda4beee28c0b80c249b5618c3a0d5e960f63929f380ac9stduhpfstephduh@live.fr2025-07-22T12:51:03+02:00GitHubnoreply@github.com2025-07-22T12:51:03+02:00Mtmd: add a way to select device for vision encoder (#14236)
1131e28c0b80c249b5618c3a0d5e960f63929f380ac98e6f8bc875358968b63e08f7bbbe0a288f29d856Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-22T12:33:10+02:00GitHubnoreply@github.com2025-07-22T12:33:10+02:00cuda : implement bf16 cpy ops and enable bf16 cont (#14763)
11328e6f8bc875358968b63e08f7bbbe0a288f29d856adef81781a15083f218eae6c488b95cdad781971lhezlih@qti.qualcomm.com2025-07-21T23:53:30-07:00GitHubnoreply@github.com2025-07-22T08:53:30+02:00opencl: remove unreachable `return` (#14806)
1133adef81781a15083f218eae6c488b95cdad78197148b86c4fdb1b1246d9fe17d2d53e3a1c2a0c3245Molly Sophiamollysophia379@gmail.com2025-07-22T09:24:22+08:00GitHubnoreply@github.com2025-07-22T09:24:22+08:00server : allow setting `--reverse-prompt` arg (#14799)
113448b86c4fdb1b1246d9fe17d2d53e3a1c2a0c324538d3af1b73302377111e88ddd725257638339286R0CKSTARyeahdongcn@gmail.com2025-07-22T07:45:26+08:00GitHubnoreply@github.com2025-07-22T07:45:26+08:00cuda: remove linking to cublasLt (#14790)
113538d3af1b73302377111e88ddd7252576383392866c9ee3b17e19dcc82ab93d52ae46fdd0226d4777Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-21T22:55:10+02:00GitHubnoreply@github.com2025-07-21T13:55:10-07:00opencl: fix `im2col` when `KW!=KH` (#14803)
11366c9ee3b17e19dcc82ab93d52ae46fdd0226d4777cd465d823c378853f1f6570eebfb77f69c7e1d39rmatifkingrealriadh@gmail.com2025-07-21T19:03:19+02:00GitHubnoreply@github.com2025-07-21T10:03:19-07:00opencl: add conv2d kernel (#14403)
1137cd465d823c378853f1f6570eebfb77f69c7e1d39922042601b8a16877ccb1c2afaa2071f76734f10Romain Biessyromain.biessy@codeplay.com2025-07-21T18:39:29+02:00GitHubnoreply@github.com2025-07-21T18:39:29+02:00sycl: Fix im2col (#14797)
1138922042601b8a16877ccb1c2afaa2071f76734f102ba1333b35e471b344974dde553db11bf1c2836fCharles Xucharles.xu@arm.com2025-07-21T15:49:52+02:00GitHubnoreply@github.com2025-07-21T16:49:52+03:00kleidiai: add support for get_rows (#14676)
11392ba1333b35e471b344974dde553db11bf1c2836fc2e058f1b4e799f1be085560c1bcef95b7b5ed02Radoslav Gerganovrgerganov@gmail.com2025-07-21T15:03:49+03:00GitHubnoreply@github.com2025-07-21T14:03:49+02:00docs : fix backends table in README.md (#14796)
1140c2e058f1b4e799f1be085560c1bcef95b7b5ed02c82d48ec23fb8749c341d0838f6891fd5f6b6da0Jeff Bolzjbolz@nvidia.com2025-07-21T06:35:40-05:00GitHubnoreply@github.com2025-07-21T13:35:40+02:00vulkan/cuda: Fix im2col when KW!=KH (#14789)
1141c82d48ec23fb8749c341d0838f6891fd5f6b6da0b4efd77f8ab407836ca73a5176f041650c5b2411Molly Sophiamollysophia379@gmail.com2025-07-21T17:38:36+08:00GitHubnoreply@github.com2025-07-21T17:38:36+08:00llama : fix `--reverse-prompt` crashing issue (#14794)
1142b4efd77f8ab407836ca73a5176f041650c5b24112be60cbc2707359241c2784f9d2e30d8fc7cdabbIsaacDynamo61521674+IsaacDynamo@users.noreply.github.com2025-07-21T09:24:51+02:00GitHubnoreply@github.com2025-07-21T10:24:51+03:00server : add parse_special option to /tokenize endpoint (#14783)
11432be60cbc2707359241c2784f9d2e30d8fc7cdabbb526ad2668944a7b2b1721f60679153646313831Aman Guptaamangupta052@gmail.com2025-07-21T02:13:47+08:00GitHubnoreply@github.com2025-07-20T20:13:47+02:00docs : fix link for tools/perplexity in README.md (#14780)
1144b526ad2668944a7b2b1721f60679153646313831938b785764683c298e7805e712f8728489cc2f18rspOverflow217881046+rspOverflow@users.noreply.github.com2025-07-20T23:55:32+07:00GitHubnoreply@github.com2025-07-20T18:55:32+02:00Documentation: Further revisions to the Vulkan section in build.md (#14785)
1145938b785764683c298e7805e712f8728489cc2f1836c153248faf969af1b62ab231348694b2047b8bAman Guptaamangupta052@gmail.com2025-07-20T19:42:34+08:00GitHubnoreply@github.com2025-07-20T19:42:34+08:00Clang-format: local files first + fix BinPacking (#14779)
114636c153248faf969af1b62ab231348694b2047b8ba979ca22db0d737af1e548a73291193655c6be990cc4mpicard12@live.de2025-07-19T22:47:21+02:00GitHubnoreply@github.com2025-07-19T23:47:21+03:00Contrib: add 0cc4m as codeowner for Vulkan backend (#14775)
1147a979ca22db0d737af1e548a73291193655c6be9990083283ec254fa8d33897746dea229aee401b37Ervin Áron Tasnádietasnadi@protonmail.com2025-07-19T21:59:08+02:00GitHubnoreply@github.com2025-07-19T21:59:08+02:00ggml: adds CONV_2D op and direct GEMM Vulkan implementation (#14316)
114890083283ec254fa8d33897746dea229aee401b37d4b91ea7b2da253e1355b503f0fcb7b428ce005dcompiladegit@compilade.net2025-07-19T12:51:22-04:00GitHubnoreply@github.com2025-07-19T12:51:22-04:00imatrix : use GGUF to store importance matrices (#9400)
1149d4b91ea7b2da253e1355b503f0fcb7b428ce005d83f5872404baa39d826af2ef66351e63c64205a8Peter0x44peter0x44@disroot.org2025-07-19T16:58:03+01:00GitHubnoreply@github.com2025-07-19T17:58:03+02:00vulkan: Add logging for bf16 features to ggml_vk_print_gpu_info (#13274) (#14707)
115083f5872404baa39d826af2ef66351e63c64205a8f0d4d176df72734a543c29eef9f942850c13311e0cc4mpicard12@live.de2025-07-19T17:47:53+02:00GitHubnoreply@github.com2025-07-19T17:47:53+02:00Vulkan: Fix fprintf format-security warning (#14770)
1151f0d4d176df72734a543c29eef9f942850c13311eb17230917c18a25af9cd143a941001466af845a2rspOverflow217881046+rspOverflow@users.noreply.github.com2025-07-19T17:18:36+07:00GitHubnoreply@github.com2025-07-19T12:18:36+02:00Documentation: Update build.md's Vulkan section (#14736)
1152b17230917c18a25af9cd143a941001466af845a2bf9087f59aab940cf312b85a67067ce33d9e365aGeorgi Gerganovggerganov@gmail.com2025-07-19T11:46:12+03:00Georgi Gerganovggerganov@gmail.com2025-07-19T11:46:50+03:00sync : ggml
1153bf9087f59aab940cf312b85a67067ce33d9e365a9fb1042ce6719bc46dbe88ab013148aabe3105f1Georgi Gerganovggerganov@gmail.com2025-07-18T20:37:26+03:00GitHubnoreply@github.com2025-07-18T20:37:26+03:00metal : fuse add, mul + add tests (#14596)
11549fb1042ce6719bc46dbe88ab013148aabe3105f12adf8d83acdb9b1bf58db6c9729ac9dc6847a58bGeorgi Gerganovggerganov@gmail.com2025-07-18T20:08:33+03:00GitHubnoreply@github.com2025-07-18T20:08:33+03:00graph : fix graph reuse reset of params (#14760)
11552adf8d83acdb9b1bf58db6c9729ac9dc6847a58b021cc28bef4dd7d0bf9c91dbbd0803caa6cb15f2Georgi Gerganovggerganov@gmail.com2025-07-18T17:33:41+03:00GitHubnoreply@github.com2025-07-18T17:33:41+03:00parallel : add option for different RNG seeds (#14757)
1156021cc28bef4dd7d0bf9c91dbbd0803caa6cb15f2d498af3d5a00f96bdd37b534860f03a6d9e98d39Oliver Simonsoliver.simons@posteo.de2025-07-18T13:35:32+02:00GitHubnoreply@github.com2025-07-18T04:35:32-07:00cuda : Fix Gemma3n not executed as CUDA_GRAPH on NVGPUs (#14741)
1157d498af3d5a00f96bdd37b534860f03a6d9e98d39eacdeb5bfcb6c6cd54461fd0e9f04cab78bf975bGeorgi Gerganovggerganov@gmail.com2025-07-18T14:31:15+03:00GitHubnoreply@github.com2025-07-18T14:31:15+03:00graph : avoid huge warm-up graphs for MoE models (#14753)
1158eacdeb5bfcb6c6cd54461fd0e9f04cab78bf975be0cb5c5cb8a61ac232130cf6bf878035f93824d9Georgi Gerganovggerganov@gmail.com2025-07-18T11:53:55+03:00GitHubnoreply@github.com2025-07-18T11:53:55+03:00model : fix build after merge conflict (#14754)
1159e0cb5c5cb8a61ac232130cf6bf878035f93824d9f9a31eea06a859e34cecb88b4d020c7f03d86cc4lgai-exaoneexaonemodels@lgresearch.ai2025-07-18T17:45:49+09:00GitHubnoreply@github.com2025-07-18T10:45:49+02:00model : add EXAONE 4.0 support (#14630)
1160f9a31eea06a859e34cecb88b4d020c7f03d86cc48f974bc1e980c06833504276021072e7a4088c81Aman Guptaamangupta052@gmail.com2025-07-18T14:54:18+08:00GitHubnoreply@github.com2025-07-18T14:54:18+08:00CUDA: set_rows + cpy.cu refactor (#14712)
11618f974bc1e980c06833504276021072e7a4088c8109651d09ffc1e941bd1be23163abf5495c416547Georgi Gerganovggerganov@gmail.com2025-07-18T08:29:28+03:00GitHubnoreply@github.com2025-07-18T08:29:28+03:00graph : refactor context to not pass gf explicitly (#14629)
116209651d09ffc1e941bd1be23163abf5495c416547349ea79fcebc75b0c55bf61594a47736966d4f95Nexes the Elder124105151+Nexesenex@users.noreply.github.com2025-07-18T06:25:54+02:00GitHubnoreply@github.com2025-07-18T07:25:54+03:00graph : Pass the graph placeholder message in debug mode (#14748)
1163349ea79fcebc75b0c55bf61594a47736966d4f95670e1360cd40f242ae76ba0966542fae6cb59392Neo Zhang Jianyujianyu.zhang@intel.com2025-07-18T10:23:14+08:00GitHubnoreply@github.com2025-07-18T10:23:14+08:00use max work group size for device to replace the magic number (#14732)
1164670e1360cd40f242ae76ba0966542fae6cb59392760b4484e3c192a2649a6ffd0d90086c5558f849Piotr Wilkin (ilintar)piotr.wilkin@syndatis.com2025-07-18T01:17:16+02:00GitHubnoreply@github.com2025-07-18T01:17:16+02:00convert : fix Ernie4.5 MoE without shared experts (#14746)
1165760b4484e3c192a2649a6ffd0d90086c5558f849cb887f1bc1001c92f7b4a595b9014f3a454a07abWroclawwroclaw223@outlook.com2025-07-18T00:18:16+02:00GitHubnoreply@github.com2025-07-17T15:18:16-07:00nix : use optionalAttrs for env mkDerivation attrset argument (#14726)
1166cb887f1bc1001c92f7b4a595b9014f3a454a07abd6fb3f6b49b27ef1c0f4cf5128e041f7e7dc03afPiotr Wilkin (ilintar)piotr.wilkin@syndatis.com2025-07-17T23:15:32+02:00GitHubnoreply@github.com2025-07-17T23:15:32+02:00model: add Ernie 4.5 MoE support (#14658)
1167d6fb3f6b49b27ef1c0f4cf5128e041f7e7dc03af01612b74090df592663cfa01f661c9628f403b59Georgi Gerganovggerganov@gmail.com2025-07-17T20:52:33+03:00GitHubnoreply@github.com2025-07-17T20:52:33+03:00kv-cache : fix k-shift for multiple streams (#14742)
116801612b74090df592663cfa01f661c9628f403b59086cf81e88fb75287b71ff19c08a206b7bc2e02fGeorgi Gerganovggerganov@gmail.com2025-07-17T19:08:33+03:00GitHubnoreply@github.com2025-07-17T19:08:33+03:00llama : reuse compute graphs (#14482)
1169086cf81e88fb75287b71ff19c08a206b7bc2e02fd9b691081c04ec5fb0daa9d2b979f915c142963dTarek Dakhrantarek@liquid.ai2025-07-17T09:22:11+02:00GitHubnoreply@github.com2025-07-17T09:22:11+02:00llama : fix parallel processing for lfm2 (#14705)
1170d9b691081c04ec5fb0daa9d2b979f915c142963dad57d3edd2f48cf6dc41a98fd9b303435ecb4fb0Georgi Gerganovggerganov@gmail.com2025-07-17T09:49:15+03:00GitHubnoreply@github.com2025-07-17T09:49:15+03:00kv-cache : opt mask set input (#14600)
1171ad57d3edd2f48cf6dc41a98fd9b303435ecb4fb01ba45d49822c39ca9a552c7b75efe0495ff400c3Georgi Gerganovggerganov@gmail.com2025-07-17T09:45:54+03:00GitHubnoreply@github.com2025-07-17T09:45:54+03:00batch : fix uninitialized has_cpl flag (#14733)
11721ba45d49822c39ca9a552c7b75efe0495ff400c319e5943d9e976b59ccd5a0a87ae3d2ff3da44390Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-17T01:52:08+02:00GitHubnoreply@github.com2025-07-16T20:52:08-03:00ci : disable failing vulkan crossbuilds (#14723)
117319e5943d9e976b59ccd5a0a87ae3d2ff3da44390496957e1cbcb522abc63aa18521036e40efce985Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-16T23:17:43+02:00GitHubnoreply@github.com2025-07-16T23:17:43+02:00convert : make hf token optional (#14717)
1174496957e1cbcb522abc63aa18521036e40efce98521c021745d781edf9c44b4972ef6cbbf53b0ecffDiner Burgerburger@diner.name2025-07-16T15:17:25-04:00GitHubnoreply@github.com2025-07-16T21:17:25+02:00llama : fix parameter order for hybrid memory initialization (#14725)
117521c021745d781edf9c44b4972ef6cbbf53b0ecffb0f0ecc3dce806c68609d375a2b3edc430d8db18Reese Levinereeselevine1@gmail.com2025-07-16T08:18:51-07:00GitHubnoreply@github.com2025-07-16T18:18:51+03:00ggml: Add initial WebGPU backend (#14521)
1176b0f0ecc3dce806c68609d375a2b3edc430d8db18225e7a1438f4ea85eaa7b5ef3ab3b266ee4d9c06tempstudio49735574+tempstudio@users.noreply.github.com2025-07-16T10:02:06-05:00GitHubnoreply@github.com2025-07-16T18:02:06+03:00model : support output bias for qwen2 (#14711)
1177225e7a1438f4ea85eaa7b5ef3ab3b266ee4d9c06ab140198211385b85eeeb0abd549a4bbe259e10dGeorgi Gerganovggerganov@gmail.com2025-07-16T16:35:42+03:00GitHubnoreply@github.com2025-07-16T16:35:42+03:00llama : add high-throughput mode (#14363)
1178ab140198211385b85eeeb0abd549a4bbe259e10d64978340b0b4a0a6e2fb74270c1509383d2eff32Aman Guptaamangupta052@gmail.com2025-07-16T20:03:51+08:00GitHubnoreply@github.com2025-07-16T20:03:51+08:00Support diffusion models: Add Dream 7B (#14644)
117964978340b0b4a0a6e2fb74270c1509383d2eff326ffd4e9c442e99afac3d138543ebf86d5fc5ee03Georgi Gerganovggerganov@gmail.com2025-07-16T14:43:32+03:00GitHubnoreply@github.com2025-07-16T14:43:32+03:00ggml : add asserts (#14720)
11806ffd4e9c442e99afac3d138543ebf86d5fc5ee03e4841d24d3485ea4af54f8b65a19ec3123f0ff3cGeorgi Gerganovggerganov@gmail.com2025-07-16T14:04:12+03:00GitHubnoreply@github.com2025-07-16T14:04:12+03:00server : pre-calculate EOG logit biases (#14721)
1181e4841d24d3485ea4af54f8b65a19ec3123f0ff3c538cc77f7f44dfa047dba6a06d90c86dda69cf1dShunta Saitoshunta.saito@gmail.com2025-07-16T19:12:22+09:00GitHubnoreply@github.com2025-07-16T12:12:22+02:00llama : fix parallel processing for plamo2 (#14716)
1182538cc77f7f44dfa047dba6a06d90c86dda69cf1d5cae76654113160f691f581930b69fc5535e8159Georgi Gerganovggerganov@gmail.com2025-07-16T12:13:57+03:00GitHubnoreply@github.com2025-07-16T12:13:57+03:00server : fix handling of the ignore_eos flag (#14710)
11835cae76654113160f691f581930b69fc5535e81594b91d6f71f14040979bcdb7b6729b3bca93ec1c1Johannes Gäßlerjohannesg@5d6.de2025-07-16T09:33:28+02:00GitHubnoreply@github.com2025-07-16T09:33:28+02:00scripts: synthetic prompt mode for server-bench.py (#14695)
11844b91d6f71f14040979bcdb7b6729b3bca93ec1c1cf91f217f1c8b98b6db8e4ba6b480f017f81d206Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-16T08:52:04+02:00GitHubnoreply@github.com2025-07-16T08:52:04+02:00convert : only check for tokenizer folder if we need it (#14704)
1185cf91f217f1c8b98b6db8e4ba6b480f017f81d20679e0b68c178656bb0632cb8602d2940b755077f8Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-16T08:51:12+02:00GitHubnoreply@github.com2025-07-16T08:51:12+02:00convert : add pre-computed hashes first to prevent order mishaps (#14701)
118679e0b68c178656bb0632cb8602d2940b755077f8c81f4192f91a1e209c1eec7a84fe5371ef9175daMin-Hua136287195+Min-Hua@users.noreply.github.com2025-07-16T12:00:42+08:00GitHubnoreply@github.com2025-07-16T07:00:42+03:00llama: add LLAMA_API to deprecated llama_kv_self_seq_div (#14708)
1187c81f4192f91a1e209c1eec7a84fe5371ef9175da4a4f426944e79b79e389f9ed7b34831cb9b637adEd Addario29247825+EAddario@users.noreply.github.com2025-07-15T23:04:42+01:00GitHubnoreply@github.com2025-07-16T00:04:42+02:00gguf-py : dump bpw per layer and model in markdown mode (#14703)
11884a4f426944e79b79e389f9ed7b34831cb9b637adba1ceb34566c889a1fc500efa79799ffed25d9b0Gabriel Larson55459720+gabriellarson@users.noreply.github.com2025-07-15T14:54:22-05:00GitHubnoreply@github.com2025-07-15T21:54:22+02:00model : add Kimi-K2 support (#14654)
1189ba1ceb34566c889a1fc500efa79799ffed25d9b010a0351a97c25471aea0bbde9cca54d32d163eecJeff Bolzjbolz@nvidia.com2025-07-15T14:51:09-05:00GitHubnoreply@github.com2025-07-15T21:51:09+02:00vulkan: fix noncontig check for mat_mul_id splitting (#14683)
119010a0351a97c25471aea0bbde9cca54d32d163eec68e37a61a7b24863f67541db65b4f6195962a268Jeff Bolzjbolz@nvidia.com2025-07-15T14:32:11-05:00GitHubnoreply@github.com2025-07-15T21:32:11+02:00vulkan: add RTE variants for glu/add/sub/mul/div (#14653)
119168e37a61a7b24863f67541db65b4f6195962a268cbc68be51d88b1d5531643b926a4b359c3cff131Shunta Saitoshunta.saito@gmail.com2025-07-16T01:11:42+09:00GitHubnoreply@github.com2025-07-15T18:11:42+02:00model : add PLaMo-2 support (#14560)
1192cbc68be51d88b1d5531643b926a4b359c3cff131bdca38376f7e8dd928defe01ce6a16218a64b040R0CKSTARyeahdongcn@gmail.com2025-07-15T15:28:53+08:00GitHubnoreply@github.com2025-07-15T15:28:53+08:00cuda: fix build warnings in set-rows.cu (unused variable) (#14687)
1193bdca38376f7e8dd928defe01ce6a16218a64b04055c509daf51d25bfaee9c8b8ce6abff103d4473bAnton Mitkovanton_b_mitkov@abv.bg2025-07-14T18:12:42+01:00GitHubnoreply@github.com2025-07-14T18:12:42+01:00sycl: Hotfix for non dnnl codepath (#14677)
119455c509daf51d25bfaee9c8b8ce6abff103d4473b9c9e4fc6354fc811efa06a8eb7a86d3315cec9c8shalinib-ibmShalini.Salomi.Bodapati@ibm.com2025-07-14T18:46:42+05:30GitHubnoreply@github.com2025-07-14T16:16:42+03:00ggml : refactor llamafile_sgemm PPC code (#14673)
11959c9e4fc6354fc811efa06a8eb7a86d3315cec9c8494c5899cb76859f32ddd913534f2685fd684a3dAman Guptaamangupta052@gmail.com2025-07-14T21:01:41+08:00GitHubnoreply@github.com2025-07-14T21:01:41+08:00llama-context: add ability to get logits (#14672)
1196494c5899cb76859f32ddd913534f2685fd684a3d0f4c6ec0f1a9607ba67071f8a02c69b0afc2f91eJohannes Gäßlerjohannesg@5d6.de2025-07-14T13:14:30+02:00GitHubnoreply@github.com2025-07-14T13:14:30+02:00scripts: benchmark for HTTP server throughput (#14668)
11970f4c6ec0f1a9607ba67071f8a02c69b0afc2f91e65a3ebb0aa56d6c501466e0f950ad15105fd32d8Akarshan Biswasakarshan@menlo.ai2025-07-14T15:07:55+05:30GitHubnoreply@github.com2025-07-14T10:37:55+01:00SYCL: use 1D kernel for set_rows (#14618)
119865a3ebb0aa56d6c501466e0f950ad15105fd32d80d9226763c82562186122f3b827fa3862864a19cAnton Mitkovanton_b_mitkov@abv.bg2025-07-14T10:37:35+01:00GitHubnoreply@github.com2025-07-14T10:37:35+01:00sycl: Batched mulmat rework for oneDNN dispatch (#14617)
11990d9226763c82562186122f3b827fa3862864a19c982e347255723fe6d02e60ee30cfdd0559c884c5Molly Sophiamollysophia379@gmail.com2025-07-14T07:43:43+08:00GitHubnoreply@github.com2025-07-14T07:43:43+08:00llama : add jinja template for rwkv-world (#14665)
1200982e347255723fe6d02e60ee30cfdd0559c884c5923e3ea2e3c96a0b4c208f53bc3bc90cdcdf13c0Ed Addario29247825+EAddario@users.noreply.github.com2025-07-13T17:02:17+01:00GitHubnoreply@github.com2025-07-13T18:02:17+02:00quantize : fix minor logic flaw in --tensor-type (#14572)
1201923e3ea2e3c96a0b4c208f53bc3bc90cdcdf13c0e743cddb60dc3a8815b9de7dd7d5c491e61b2259Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-13T15:01:24+02:00GitHubnoreply@github.com2025-07-13T15:01:24+02:00cuda : add set rows for bf16 (#14664)
1202e743cddb60dc3a8815b9de7dd7d5c491e61b225905fec5bd298d3c0243cbb9336e59b8b6aff75a81Yavor Ivanovyavorgenadiev@gmail.com2025-07-13T02:33:16-07:00GitHubnoreply@github.com2025-07-13T11:33:16+02:00cuda : add ELU support (#14657)
120305fec5bd298d3c0243cbb9336e59b8b6aff75a81dcf7f2ea3c4e3cf36dd9ab5a36785c00e6033267Georgi Gerganovggerganov@gmail.com2025-07-13T10:36:33+03:00GitHubnoreply@github.com2025-07-13T10:36:33+03:00ggml : add build-time message to remind about ggml_set_rows (#14661)
1204dcf7f2ea3c4e3cf36dd9ab5a36785c00e603326784b396e0510855a95d591afdf1f21c562cb3712aYavor Ivanovyavorgenadiev@gmail.com2025-07-12T22:38:13-07:00GitHubnoreply@github.com2025-07-13T08:38:13+03:00metal : Add missing unary ops Metal support (#14660)
120584b396e0510855a95d591afdf1f21c562cb3712ac31e60647def83d671bac5ab5b35579bf25d9aa1Yavor Ivanovyavorgenadiev@gmail.com2025-07-12T22:12:36-07:00GitHubnoreply@github.com2025-07-13T08:12:36+03:00cmake : Add CMake presets for Linux and GCC (#14656)
1206c31e60647def83d671bac5ab5b35579bf25d9aa167eade1bf93f40e7c4975e5bd0782f426c89cff4Tarek Dakhrantarek@liquid.ai2025-07-12T19:10:14+02:00GitHubnoreply@github.com2025-07-12T19:10:14+02:00tests : cover lfm2 cases in test_ssm_conv (#14651)
120767eade1bf93f40e7c4975e5bd0782f426c89cff47de5c7cab61d4da4387ed9b216f88b96297bcc2dTarek Dakhrantarek@liquid.ai2025-07-12T19:07:08+02:00GitHubnoreply@github.com2025-07-12T19:07:08+02:00docs : add LFM2 to models section (#14650)
12087de5c7cab61d4da4387ed9b216f88b96297bcc2d8eff95544e817704d44bec20f9fc956ce76a33beAman Guptaamangupta052@gmail.com2025-07-12T21:31:38+08:00GitHubnoreply@github.com2025-07-12T16:31:38+03:00CUDA: add set rows for f32 and f16 (#14551)
12098eff95544e817704d44bec20f9fc956ce76a33be3120413ccd7797d4cbdee32ef89a641765d1f6c4Georgi Gerganovggerganov@gmail.com2025-07-12T16:06:12+03:00Georgi Gerganovggerganov@gmail.com2025-07-12T16:13:27+03:00sync : ggml
12103120413ccd7797d4cbdee32ef89a641765d1f6c4215535701d659e873c52d2a9163d4616a42da4f7Georgi Gerganovggerganov@gmail.com2025-07-12T12:39:32+03:00Georgi Gerganovggerganov@gmail.com2025-07-12T14:25:44+03:00vulkan : remove unused vars (#0)
1211215535701d659e873c52d2a9163d4616a42da4f774bb294591054a6bf0cc3f6e4637d87414cf2c46Georgi Gerganovggerganov@gmail.com2025-07-12T12:39:27+03:00Georgi Gerganovggerganov@gmail.com2025-07-12T14:25:44+03:00sync : ggml
121274bb294591054a6bf0cc3f6e4637d87414cf2c463e303b1107e4748cd50a675544d8c7fabb40ec54Aclyaclysia@gmail.com2025-07-12T12:37:37+03:00Georgi Gerganovggerganov@gmail.com2025-07-12T14:25:44+03:00vulkan : implement bilinear interpolation (ggml/1291)
12133e303b1107e4748cd50a675544d8c7fabb40ec540c1df14b5f8d992805cb22d0b77b44092a18aeabAclyaclysia@gmail.com2025-07-12T12:32:32+03:00Georgi Gerganovggerganov@gmail.com2025-07-12T14:25:44+03:00vulkan : implement ggml_roll (ggml/1290)
12140c1df14b5f8d992805cb22d0b77b44092a18aeabb3ad3a0191994d6c47b2bd389d5c7431526ecd2cDouglas Hanleythesecretaryofwar@gmail.com2025-07-12T06:21:02-04:00GitHubnoreply@github.com2025-07-12T13:21:02+03:00server : fix pooled embedding output (#14645)
1215b3ad3a0191994d6c47b2bd389d5c7431526ecd2c98197e5c98388470030d908f355ec5937dcccaaaJeff Bolzjbolz@nvidia.com2025-07-12T05:12:26-05:00GitHubnoreply@github.com2025-07-12T12:12:26+02:00vulkan: support SET_ROWS (#14587)
121698197e5c98388470030d908f355ec5937dcccaaaf5e96b368f1acc7f53c390001b936517c4d18999Jeff Bolzjbolz@nvidia.com2025-07-12T04:51:58-05:00GitHubnoreply@github.com2025-07-12T11:51:58+02:00vulkan: optimizations for deepseek prompt processing (#14555)
1217f5e96b368f1acc7f53c390001b936517c4d18999756aa1020aabc70b71b9a8efcd35fbf42b3bb9abTarek Dakhrant.dakhran@gmail.com2025-07-11T20:27:01+02:00GitHubnoreply@github.com2025-07-11T20:27:01+02:00model : support LiquidAI LFM2 hybrid family (#14620)
1218756aa1020aabc70b71b9a8efcd35fbf42b3bb9abaaa088d87f9006d56866085fe46e4b2755ef723fSlobodan Josic127323561+slojosic-amd@users.noreply.github.com2025-07-11T18:55:00+02:00GitHubnoreply@github.com2025-07-11T18:55:00+02:00HIP : Add HIP 7.0+ compatibility for hipBLAS compute types (#14634)
1219aaa088d87f9006d56866085fe46e4b2755ef723f0d5375d54b258ec63edd1fb5d58c37d58ce8be8bGeorgi Gerganovggerganov@gmail.com2025-07-11T16:07:55+03:00GitHubnoreply@github.com2025-07-11T16:07:55+03:00readme : add hot PRs (#14636)
12200d5375d54b258ec63edd1fb5d58c37d58ce8be8b576c82eda210ca0111c04f5256bf77897a4d4cc4Georgi Gerganovggerganov@gmail.com2025-07-11T13:46:07+03:00GitHubnoreply@github.com2025-07-11T13:46:07+03:00llama : move enum llama_vocab_pre_type to implementation (#14631)
1221576c82eda210ca0111c04f5256bf77897a4d4cc40aedae00e6fb48680324a5ac5da9cba0e35de6b5Dowonks2515@naver.com2025-07-11T16:36:04+09:00GitHubnoreply@github.com2025-07-11T09:36:04+02:00vocab : add midm-2.0 model pre-tokenizer (#14626)
12220aedae00e6fb48680324a5ac5da9cba0e35de6b56bdda13981d6c8189b7dc4f9fb8ecb91c21529f8Gabe Goodhartghart@us.ibm.com2025-07-10T18:20:13-06:00GitHubnoreply@github.com2025-07-11T02:20:13+02:00model : Granite Four (#13550)
12236bdda13981d6c8189b7dc4f9fb8ecb91c21529f80b8855775c6b873931d40b77a5e42558aacbde52rmatifkingrealriadh@gmail.com2025-07-10T23:58:12+02:00GitHubnoreply@github.com2025-07-10T14:58:12-07:00opencl: add tiled mul_mat_f16_f32 (#14535)
12240b8855775c6b873931d40b77a5e42558aacbde524bb625b713fd9b294b4f7af87eaa752b710c7cc1lhezquic_lih@quicinc.com2025-07-10T11:48:52-07:00GitHubnoreply@github.com2025-07-10T11:48:52-07:00opencl: add `set_rows` for `f16` and `f32` (#14547)
12254bb625b713fd9b294b4f7af87eaa752b710c7cc111ee0fea2a24da1d3206eeba8fc52b759d9dfb24Ryan Mangeno160974989+ryan-mangeno@users.noreply.github.com2025-07-10T13:41:00-04:00GitHubnoreply@github.com2025-07-10T19:41:00+02:00Smoldocling support (#14597)
122611ee0fea2a24da1d3206eeba8fc52b759d9dfb24a457551332853ef19d0796fec12b62c538126ea5Aman Guptaamangupta052@gmail.com2025-07-10T23:29:01+08:00GitHubnoreply@github.com2025-07-10T23:29:01+08:00Docs: script to auto-generate ggml operations docs (#14598)
1227a457551332853ef19d0796fec12b62c538126ea5704bb7a71c01dc07c1478b85f6322bf5dfde1eafEric Zhang34133756+EZForever@users.noreply.github.com2025-07-10T20:29:05+08:00GitHubnoreply@github.com2025-07-10T15:29:05+03:00cmake : do not search for curl libraries by ourselves (#14613)
1228704bb7a71c01dc07c1478b85f6322bf5dfde1eaf435a6d10d618a015060e45a38c7e9f27f4243316Akarshan Biswasakarshan@menlo.ai2025-07-10T13:59:38+05:30GitHubnoreply@github.com2025-07-10T09:29:38+01:00SYCL: Initial set_rows kernel implementation (#14562)
1229435a6d10d618a015060e45a38c7e9f27f4243316f9a867f5921a85f3fa64d7b067f4c8ffc5f62eb4Xuan-Son Nguyenson@huggingface.co2025-07-10T09:00:20+02:00GitHubnoreply@github.com2025-07-10T10:00:20+03:00llama : minor coding style fix for smollm3 (#14605)
1230f9a867f5921a85f3fa64d7b067f4c8ffc5f62eb4ac44eb6c808bd5d677261ce86edd8c43ec54cf2cEric Zhang34133756+EZForever@users.noreply.github.com2025-07-10T13:19:37+08:00GitHubnoreply@github.com2025-07-10T08:19:37+03:00cmake : bump llguidance version to v1.0.1 (#14609)
1231ac44eb6c808bd5d677261ce86edd8c43ec54cf2ca57d1bcb3c0165ac87b1f0dbb429839b0da69689Eric Zhang34133756+EZForever@users.noreply.github.com2025-07-10T13:19:13+08:00GitHubnoreply@github.com2025-07-10T08:19:13+03:00cmake : llguidance build parser library only (#14608)
1232a57d1bcb3c0165ac87b1f0dbb429839b0da69689cb9178f885d1986cc0b12feb26ff426bc8a3556ccompiladegit@compilade.net2025-07-09T23:54:38-04:00GitHubnoreply@github.com2025-07-09T23:54:38-04:00cuda : support Falcon-H1 state size for SSM_SCAN (#14602)
1233cb9178f885d1986cc0b12feb26ff426bc8a3556c4a5686da22057867c23bd4a6be941ddc8c51e585Xuan-Son Nguyenson@huggingface.co2025-07-09T23:09:28+02:00GitHubnoreply@github.com2025-07-09T23:09:28+02:00llama : remove llm_graph_input_one (#14603)
12344a5686da22057867c23bd4a6be941ddc8c51e58598bab638fb28cf95a5a66dd2d51b40d6c8f6d69acompiladegit@compilade.net2025-07-09T14:59:57-04:00GitHubnoreply@github.com2025-07-09T14:59:57-04:00llama : support Jamba hybrid Transformer-Mamba models (#7531)
123598bab638fb28cf95a5a66dd2d51b40d6c8f6d69a26a48ad699d50b6268900062661bd22f3e792579Xuan-Son Nguyenson@huggingface.co2025-07-09T18:16:12+02:00GitHubnoreply@github.com2025-07-09T18:16:12+02:00ggml : add ggml_scale_bias (#14417)
123626a48ad699d50b6268900062661bd22f3e792579ffd59e7d18a76459d5c31ba97073c7c9d73cb752Miaoqian Linlinmq006@gmail.com2025-07-09T20:33:53+08:00GitHubnoreply@github.com2025-07-09T14:33:53+02:00ggml : prevent integer overflow in gguf tensor size calculation (#14595)
1237ffd59e7d18a76459d5c31ba97073c7c9d73cb752105554595f9a7bf3e02232ed7798201d47c2a4a2Dowonks2515@naver.com2025-07-09T17:22:31+09:00GitHubnoreply@github.com2025-07-09T11:22:31+03:00model : add skt/A.X-4.0 model vocabulary (#14589)
1238105554595f9a7bf3e02232ed7798201d47c2a4a204655063c47af6cbced295c8c7ad369402b15300Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-09T10:19:50+02:00GitHubnoreply@github.com2025-07-09T10:19:50+02:00llama : remove unintended whitespace (#14592)
123904655063c47af6cbced295c8c7ad369402b1530020b7bf8a32259ad9189a4797fd3e3a859c537b99ibrahim khadraoui132432132+ibrahimkhadraoui@users.noreply.github.com2025-07-09T12:03:49+04:00GitHubnoreply@github.com2025-07-09T10:03:49+02:00model : add support for Falcon-H1 family (#14534)
124020b7bf8a32259ad9189a4797fd3e3a859c537b996efcd65945a98cf6883cdd9de4c8ccd8c79d219aXuan-Son Nguyenson@huggingface.co2025-07-09T08:26:13+02:00GitHubnoreply@github.com2025-07-09T09:26:13+03:00convert : fix smollm3 jinja template (#14586)
12416efcd65945a98cf6883cdd9de4c8ccd8c79d219a699f4392a33f57c3352cf8d60bdc53db7ca235e7Jeff Bolzjbolz@nvidia.com2025-07-08T13:11:42-05:00GitHubnoreply@github.com2025-07-08T20:11:42+02:00vulkan: optimize flash attention split_k_reduce (#14554)
1242699f4392a33f57c3352cf8d60bdc53db7ca235e708382869a2d6dca5d84710f2f82cc17a9696585astevenkuangstevenkuang@tencent.com2025-07-09T00:29:29+08:00GitHubnoreply@github.com2025-07-08T18:29:29+02:00model : fix hunyuan moe chat template (#14584)
124308382869a2d6dca5d84710f2f82cc17a9696585abb4f7a9e4eec171fecf0f640b1337a1c24485560Xuan-Son Nguyenson@huggingface.co2025-07-08T18:07:01+02:00GitHubnoreply@github.com2025-07-08T18:07:01+02:00model : add SmolLM3 (#14581)
1244bb4f7a9e4eec171fecf0f640b1337a1c24485560b8eeb8741d4483daf576498cf90537b4de71633ccompiladegit@compilade.net2025-07-08T11:37:47-04:00GitHubnoreply@github.com2025-07-08T18:37:47+03:00memory : fix broken batch splits for recurrent cache (#14575)
1245b8eeb8741d4483daf576498cf90537b4de71633c17a1f0d2d407040ee242e18dd79be8bb212cfcefJeff Bolzjbolz@nvidia.com2025-07-08T08:21:21-05:00GitHubnoreply@github.com2025-07-08T15:21:21+02:00vulkan : fix rope with partial rotation and non-cont src (#14582)
124617a1f0d2d407040ee242e18dd79be8bb212cfcef8f22dc0a53338c629c1ef8fa878d8e39bfe627c9Alawode Oluwandabiradabiraalawode@yahoo.com2025-07-08T11:47:33+03:00GitHubnoreply@github.com2025-07-08T11:47:33+03:00server: Add ability to mount server at prefix (#14544)
12478f22dc0a53338c629c1ef8fa878d8e39bfe627c953903ae6fa5f1caf187889c839cdd1ad25da4018Xuan-Son Nguyenson@huggingface.co2025-07-08T10:24:06+02:00GitHubnoreply@github.com2025-07-08T11:24:06+03:00model : add hunyuan moe (#14425)
124853903ae6fa5f1caf187889c839cdd1ad25da40184d0dcd4a06080e796e6742a88f2ffa7fc41b28b8Jeff Bolzjbolz@nvidia.com2025-07-08T02:38:31-05:00GitHubnoreply@github.com2025-07-08T09:38:31+02:00vulkan: increase timeout for CI (#14574)
12494d0dcd4a06080e796e6742a88f2ffa7fc41b28b875c91de6e955d5b8f3f28173f5040593e1964eb3Georgi Gerganovggerganov@gmail.com2025-07-08T10:15:21+03:00GitHubnoreply@github.com2025-07-08T10:15:21+03:00cuda : fix rope with partial rotation and non-cont src (#14580)
125075c91de6e955d5b8f3f28173f5040593e1964eb368155c66f0e76680f34442247e589a090add22d3Aman Guptaamangupta052@gmail.com2025-07-08T10:11:18+08:00GitHubnoreply@github.com2025-07-08T10:11:18+08:00CUDA: add bilinear interpolation for upscale (#14563)
125168155c66f0e76680f34442247e589a090add22d3e1a7059053a9b8958f2d57a21fc46dbc7fb24f8eR0CKSTARyeahdongcn@gmail.com2025-07-08T07:58:30+08:00GitHubnoreply@github.com2025-07-08T07:58:30+08:00musa: fix build warnings (unused variable) (#14561)
1252e1a7059053a9b8958f2d57a21fc46dbc7fb24f8e12f55c302b35cfe900b84c5fe67c262026af9c44Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-07T23:35:35+02:00GitHubnoreply@github.com2025-07-07T23:35:35+02:00llama : fix incorrect minicpm3 v_states shape (#14571)
125312f55c302b35cfe900b84c5fe67c262026af9c44b9c3eefde1b67104bd993485ff38dd62abe9d70cSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-07T21:35:08+02:00GitHubnoreply@github.com2025-07-07T21:35:08+02:00llama : remove ggml_cont where possible (#14568)
1254b9c3eefde1b67104bd993485ff38dd62abe9d70c6491d6e4f1caf0ad2221865b4249ae6938a6308cAman Guptaamangupta052@gmail.com2025-07-07T21:45:43+08:00GitHubnoreply@github.com2025-07-07T21:45:43+08:00CUDA: add bf16 and i32 to getrows (#14529)
12556491d6e4f1caf0ad2221865b4249ae6938a6308ce592be15756ae546ba87bb5a5250b71248121971Eve139727413+netrunnereve@users.noreply.github.com2025-07-06T10:29:36ZGitHubnoreply@github.com2025-07-06T12:29:36+02:00vulkan: increase LOAD_VEC_A to 8 (IQ1/IQ2) or 4 (IQ3) (#14485)
1256e592be15756ae546ba87bb5a5250b71248121971a0374a67e2924f2e845cdc59dd67d9a44065a89cJeff Bolzjbolz@nvidia.com2025-07-06T03:08:16-05:00GitHubnoreply@github.com2025-07-06T10:08:16+02:00vulkan: fix rms_norm+mul fusion (#14545)
1257a0374a67e2924f2e845cdc59dd67d9a44065a89cddef99522d1ba74193b7394e803fab8db5c78baeJeff Bolzjbolz@nvidia.com2025-07-05T02:26:04-05:00GitHubnoreply@github.com2025-07-05T09:26:04+02:00vulkan: Handle updated FA dim2/3 definition (#14518)
1258ddef99522d1ba74193b7394e803fab8db5c78bae668168814601d2aef6161ce49ab186bc74177ae7Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-05T09:17:14+02:00GitHubnoreply@github.com2025-07-05T09:17:14+02:00server : fix assistant prefilling when content is an array (#14360)
1259668168814601d2aef6161ce49ab186bc74177ae7bac8bed248d15419137c5bc7f834582397baaebcSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-05T08:24:56+02:00GitHubnoreply@github.com2025-07-04T23:24:56-07:00opencl: add GELU_ERF (#14476)
1260bac8bed248d15419137c5bc7f834582397baaebcb81510a7b78f7c5a6f069c4f3d0e569b9d287913Georgi Gerganovggerganov@gmail.com2025-07-05T07:18:09+03:00GitHubnoreply@github.com2025-07-05T07:18:09+03:00eval-callback : check for empty input (#14539)
1261b81510a7b78f7c5a6f069c4f3d0e569b9d287913ef797db357e44ecb7437fa9d22f4e1614104b342R0CKSTARyeahdongcn@gmail.com2025-07-05T12:10:53+08:00GitHubnoreply@github.com2025-07-05T12:10:53+08:00test-backend-ops: add support for specifying output format (#14368)
1262ef797db357e44ecb7437fa9d22f4e1614104b34267d1ef23c68e1e09dc6d7423d1ef5fc0ea56ed5eGeorgi Gerganovggerganov@gmail.com2025-07-04T19:19:09+03:00GitHubnoreply@github.com2025-07-04T19:19:09+03:00metal : disable fast math in all quantize kernels (#14528)
126367d1ef23c68e1e09dc6d7423d1ef5fc0ea56ed5e7b50f7c0257695d0f6918bffce4e68d5c13b0c9eGeorgi Gerganovggerganov@gmail.com2025-07-04T09:08:59+03:00GitHubnoreply@github.com2025-07-04T09:08:59+03:00batch : add optional for sequential equal split (#14511)
12647b50f7c0257695d0f6918bffce4e68d5c13b0c9ec79184d2d192489e3c918bab8ed717d22f8c02bdGeorgi Gerganovggerganov@gmail.com2025-07-04T09:05:36+03:00GitHubnoreply@github.com2025-07-04T09:05:36+03:00graph : prepare for 4D mask (#14515)
1265c79184d2d192489e3c918bab8ed717d22f8c02bd499a8f5a787f5fdc7f3fdfb9d1ff01b6cb5e994eGeorgi Gerganovggerganov@gmail.com2025-07-04T09:04:59+03:00GitHubnoreply@github.com2025-07-04T09:04:59+03:00batch : add n_used count (#14512)
1266499a8f5a787f5fdc7f3fdfb9d1ff01b6cb5e994e28657a8229b5adc6028cf1c4ed62191792d2fdb0luyhcsu110711054+luyhcsu@users.noreply.github.com2025-07-04T11:50:07+08:00GitHubnoreply@github.com2025-07-04T11:50:07+08:00CANN: Replace aclrtMemsetSync with aclnnInplaceZero operator (#14002)
126728657a8229b5adc6028cf1c4ed62191792d2fdb0bee28421be25fd447f61cb6db64d556cbfce32ecSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-03T23:07:22+02:00GitHubnoreply@github.com2025-07-03T23:07:22+02:00ggml : implement GEGLU_ERF and GEGLU_QUICK ops (#14445)
1268bee28421be25fd447f61cb6db64d556cbfce32ec2b72bedec198a90bb5b0cceaf1d0aff9e34ffbc2lhezquic_lih@quicinc.com2025-07-03T11:22:24-07:00GitHubnoreply@github.com2025-07-03T20:22:24+02:00opencl : broadcast for soft_max (#14510)
12692b72bedec198a90bb5b0cceaf1d0aff9e34ffbc2c8c4495b8d3a8799e2d46778f993965b0ac1ae43Jeff Bolzjbolz@nvidia.com2025-07-03T13:21:14-05:00GitHubnoreply@github.com2025-07-03T20:21:14+02:00vulkan: support mixed/deepseekR1 FA head sizes (#14509)
1270c8c4495b8d3a8799e2d46778f993965b0ac1ae437b63a71a6b0f54effe9b94073d4d0519dcf53676Johannes Gäßlerjohannesg@5d6.de2025-07-03T17:05:18+02:00GitHubnoreply@github.com2025-07-03T17:05:18+02:00ggml: backward pass for split swiglu (#14483)
12717b63a71a6b0f54effe9b94073d4d0519dcf536760c2ee38ab729f3f6a7c28dc4c5dcd8d5f0cbf8fcNicolò Scipionenicolo.scipione@codeplay.com2025-07-03T11:00:03+02:00GitHubnoreply@github.com2025-07-03T11:00:03+02:00Fix conditional enabling following arch checks for ggml-sycl (#14504)
12720c2ee38ab729f3f6a7c28dc4c5dcd8d5f0cbf8fca70c8a0c4b4c1606cd9a0ba889ce61aa88610095Xuan-Son Nguyenson@huggingface.co2025-07-03T10:03:06+02:00GitHubnoreply@github.com2025-07-03T10:03:06+02:00convert : correct gemma 3n conversion (#14450)
1273a70c8a0c4b4c1606cd9a0ba889ce61aa886100959067487c4411efb20400103fcccfdd389c80d428Georgi Gerganovggerganov@gmail.com2025-07-03T10:53:35+03:00GitHubnoreply@github.com2025-07-03T10:53:35+03:00kv-cache : use ggml_set_rows (#14285)
12749067487c4411efb20400103fcccfdd389c80d428d4cdd9c1c3cebdafca735958597de4ff7b7c0f54Georgi Gerganovggerganov@gmail.com2025-07-03T10:46:57+03:00GitHubnoreply@github.com2025-07-03T10:46:57+03:00ggml : fix FA mask dim 2 and 3 (#14505)
1275d4cdd9c1c3cebdafca735958597de4ff7b7c0f5455c2646b452419be7552157b9b11373ccc1bf2f2Georgi Gerganovggerganov@gmail.com2025-07-03T07:48:32+03:00GitHubnoreply@github.com2025-07-03T07:48:32+03:00ggml : remove kompute backend (#14501)
127655c2646b452419be7552157b9b11373ccc1bf2f2e75ba4c0434eb759eb7ff74e034ebe729053e575Aman Guptaamangupta052@gmail.com2025-07-03T07:45:11+08:00GitHubnoreply@github.com2025-07-03T07:45:11+08:00CUDA: add dynamic shared mem to softmax, refactor general usage (#14497)
1277e75ba4c0434eb759eb7ff74e034ebe729053e5755d46babdc2d4675d96ebcf23cac098a02f0d30ccSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-02T21:02:35+02:00GitHubnoreply@github.com2025-07-02T21:02:35+02:00gguf-py : add support for chat template jinja files (#14508)
12785d46babdc2d4675d96ebcf23cac098a02f0d30cce17991c466ac835b2c71bd813c1ca7ff8dd97b94compiladegit@compilade.net2025-07-02T13:10:24-04:00GitHubnoreply@github.com2025-07-02T13:10:24-04:00llama : initial Mamba-2 support (#9126)
1279e17991c466ac835b2c71bd813c1ca7ff8dd97b94c46944aa25b5560e6195f8cbcbc8947e9a6395c3Georgi Gerganovggerganov@gmail.com2025-07-02T19:35:47+03:00Georgi Gerganovggerganov@gmail.com2025-07-02T20:08:45+03:00sync : ggml
1280c46944aa25b5560e6195f8cbcbc8947e9a6395c3f3ed38d793fab9f97987dc52d7a41622f2056701Daniel Beveniusdaniel.bevenius@gmail.com2025-07-02T13:55:32+02:00Georgi Gerganovggerganov@gmail.com2025-07-02T20:08:45+03:00ggml : add version function to get lib version (ggml/1286)
128155a1c5a5fdefef808e95aabd3d5563af1068cc8012a81af45f0dbbab24bd819a15f57c03ceb1be90Aman Guptaamangupta052@gmail.com2025-07-02T20:34:24+08:00Georgi Gerganovggerganov@gmail.com2025-07-02T15:48:33+03:00CUDA: add softmax broadcast (#14475)
128212a81af45f0dbbab24bd819a15f57c03ceb1be908875523eb311cac832bfda0c581e852292185ae9Johannes Gäßlerjohannesg@5d6.de2025-07-02T13:42:12+02:00Georgi Gerganovggerganov@gmail.com2025-07-02T15:48:33+03:00CUDA: broadcasting for FlashAttention mask (#14500)
12838875523eb311cac832bfda0c581e852292185ae9ec68e84c32325a3417fbcd2e60d4bda6adb4e4bcJeff Bolzjbolz@nvidia.com2025-07-01T03:32:56-05:00Georgi Gerganovggerganov@gmail.com2025-07-02T15:48:33+03:00vulkan: support softmax/FA batch and broadcast (#14449)
1284ec68e84c32325a3417fbcd2e60d4bda6adb4e4bc307e79d33d4cdd9f1d6c42fc861724e6ba12b98fGeorgi Gerganovggerganov@gmail.com2025-06-27T21:50:57+03:00Georgi Gerganovggerganov@gmail.com2025-07-02T15:48:33+03:00ggml : support bcast ggml_soft_max_ext, ggml_flash_attn_ext (#14435)
1285307e79d33d4cdd9f1d6c42fc861724e6ba12b98fd7f5f4e578d1f60b0835d1734a50438c309b3e5czhouwgzhouwg2000@gmail.com2025-07-02T20:38:10+08:00GitHubnoreply@github.com2025-07-02T14:38:10+02:00opencl : fix possible buffer overflow in dump_tensor (#14490)
1286d7f5f4e578d1f60b0835d1734a50438c309b3e5cc8a4e470f65c3a932e18eddc5fba1844876d7463Georgi Gerganovggerganov@gmail.com2025-07-02T14:12:07+03:00GitHubnoreply@github.com2025-07-02T14:12:07+03:00simple-chat : fix context-exceeded condition (#14494)
1287c8a4e470f65c3a932e18eddc5fba1844876d7463603e43dc913f90d9c3291624728b731687eddc35Eric Zhang34133756+EZForever@users.noreply.github.com2025-07-02T19:00:04+08:00GitHubnoreply@github.com2025-07-02T13:00:04+02:00opencl : skip empty nodes on cgraph compute (#14491)
1288603e43dc913f90d9c3291624728b731687eddc35611ba4b264c8fbb9d2d978bd6a5c17aeab1700fblhezquic_lih@quicinc.com2025-07-02T00:07:42-07:00GitHubnoreply@github.com2025-07-02T09:07:42+02:00opencl : update upscale to support align corners (#14488)
1289611ba4b264c8fbb9d2d978bd6a5c17aeab1700fb85841e121db5b96ae4d277a3cd3995eef66b98ecSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-02T09:02:51+02:00GitHubnoreply@github.com2025-07-02T09:02:51+02:00ci : add OpenCL to labeler workflow (#14496)
129085841e121db5b96ae4d277a3cd3995eef66b98ec68b3cd65141586ef13a698baa9029627f038f102Eric Zhang34133756+EZForever@users.noreply.github.com2025-07-02T13:41:35+08:00GitHubnoreply@github.com2025-07-02T08:41:35+03:00github : add OpenCL backend to issue templates (#14492)
129168b3cd65141586ef13a698baa9029627f038f102de569441470332ff922c23fb0413cc957be75b25Björn Ganstermail@bjoern-ganster.de2025-07-02T07:19:31+02:00GitHubnoreply@github.com2025-07-02T08:19:31+03:00ggml : Callback before abort (#14481)
1292de569441470332ff922c23fb0413cc957be75b251b2aaf28acd632b79bc3b07acb5dad877dc7dbb1Georgi Gerganovggerganov@gmail.com2025-07-01T18:04:08+03:00GitHubnoreply@github.com2025-07-01T18:04:08+03:00ci : disable fast-math for Metal GHA CI (#14478)
12931b2aaf28acd632b79bc3b07acb5dad877dc7dbb1343b6e94b61674ac94e64ede7b7ea3365793f7edGrzegorz Graszaxek@redhat.com2025-07-01T15:44:11+02:00GitHubnoreply@github.com2025-07-01T15:44:11+02:00Add Vulkan images to docker.md (#14472)
1294343b6e94b61674ac94e64ede7b7ea3365793f7ed6a746cf9c40f8d93d13eb8a6c2b989a15e7fa61aChenguang Li757486878@qq.com2025-07-01T16:47:30+08:00GitHubnoreply@github.com2025-07-01T16:47:30+08:00CANN: update aclnnGroupedMatmulV2 to aclnnGroupedMatmulV3 (#14411)
12956a746cf9c40f8d93d13eb8a6c2b989a15e7fa61aeff5e45443a248f89cc61e64786f38b68b4da489Jeff Bolzjbolz@nvidia.com2025-07-01T03:43:08-05:00GitHubnoreply@github.com2025-07-01T10:43:08+02:00vulkan: Split large mul_mat_id to fit in shared memory (#14451)
1296eff5e45443a248f89cc61e64786f38b68b4da489a6a47958a1d9edc628cedc2f9320c84b59fe1f4fSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-07-01T10:14:21+02:00GitHubnoreply@github.com2025-07-01T10:14:21+02:00add GELU_ERF (#14455)
1297a6a47958a1d9edc628cedc2f9320c84b59fe1f4ff61c05d4b1f8ad498a5cf7f0f57a40383aad563cGeorgi Gerganovggerganov@gmail.com2025-07-01T11:05:48+03:00Georgi Gerganovggerganov@gmail.com2025-07-01T11:06:39+03:00ggml : remove trailing whitespace (#0)
1298f61c05d4b1f8ad498a5cf7f0f57a40383aad563c431b2c24f31bf5411786c378b409d0202df8d53cGeorgi Gerganovggerganov@gmail.com2025-07-01T10:27:52+03:00Georgi Gerganovggerganov@gmail.com2025-07-01T11:06:39+03:00sync : ggml
1299497be7c01dab243100f9c42e0a763a746e42d03879b33b231774d5c39c8df018e9a276becae6d41aDaniel Beveniusdaniel.bevenius@gmail.com2025-06-24T06:10:16+02:00Georgi Gerganovggerganov@gmail.com2025-07-01T11:06:39+03:00ggml-quants : rename best_mad to best_error (ggml/1283)
130079b33b231774d5c39c8df018e9a276becae6d41a0a5a3b5cdfd887cf0f8e09d9ff89dee130cfcddelhezquic_lih@quicinc.com2025-07-01T00:19:16-07:00GitHubnoreply@github.com2025-07-01T09:19:16+02:00opencl : add GEGLU, REGLU, SWIGLU (#14456)
13010a5a3b5cdfd887cf0f8e09d9ff89dee130cfcdde745f11fed09ba2303f3f494efb12286d382608e5Aman Guptaamangupta052@gmail.com2025-06-30T23:57:04+08:00GitHubnoreply@github.com2025-06-30T23:57:04+08:00Add Conv2d for CPU (#14388)
1302745f11fed09ba2303f3f494efb12286d382608e55dd942de5922a22ec8446a4ad2203738dbcb9389Georgi Gerganovggerganov@gmail.com2025-06-30T18:03:03+03:00GitHubnoreply@github.com2025-06-30T18:03:03+03:00memory : correctly handle failure in apply() (#14438)
13035dd942de5922a22ec8446a4ad2203738dbcb9389a7417f55945eb7c7a5ea6807e66564b8066a4e50Georgi Gerganovggerganov@gmail.com2025-06-30T17:04:05+03:00GitHubnoreply@github.com2025-06-30T17:04:05+03:00metal : disable fast-math for some cpy kernels (#14460)
1304a7417f55945eb7c7a5ea6807e66564b8066a4e50eb3fa2913e0ace38912011fcb331624c679656f5Romain Biessyromain.biessy@codeplay.com2025-06-30T14:52:02+02:00GitHubnoreply@github.com2025-06-30T14:52:02+02:00ggml-cpu: sycl: Re-enable exp f16 (#14462)
1305eb3fa2913e0ace38912011fcb331624c679656f5c839a2da1a0d4275c3a98f70c3a7627487573a46Diego Devesaslarengh@gmail.com2025-06-30T03:43:15-07:00GitHubnoreply@github.com2025-06-30T12:43:15+02:00test-backend-ops : disable llama test (#14461)
1306c839a2da1a0d4275c3a98f70c3a7627487573a46e9b6350e61d592634263a14b3d77ecbf6c1fb096xiaobing31871554036+xiaobing318@users.noreply.github.com2025-06-30T17:48:24+08:00GitHubnoreply@github.com2025-06-30T12:48:24+03:00cmake : Remove redundant include path in CMakeLists.txt (#14452)
1307e9b6350e61d592634263a14b3d77ecbf6c1fb096caf5681fcb47dfe9bafee94ef9aa8f669ac986c7Vedran Miletićvedran@miletic.net2025-06-30T10:17:18+02:00GitHubnoreply@github.com2025-06-30T10:17:18+02:00scripts : make the shell scripts cross-platform (#14341)
1308caf5681fcb47dfe9bafee94ef9aa8f669ac986c783790b0e7e09ab17238b16452a33053a71dbdfadmatteomatteo.serva@gmail.com2025-06-29T20:02:53+02:00GitHubnoreply@github.com2025-06-29T20:02:53+02:00server : support jinja extra template kwargs (Qwen3 enable_thinking feature), from command line and from client (#13196)
130983790b0e7e09ab17238b16452a33053a71dbdfadf47c1d7106e49062279bcc57fc1077c0db61e278Renatrntk@users.noreply.github.com2025-06-29T19:29:57+02:00GitHubnoreply@github.com2025-06-29T19:29:57+02:00server : fix appearance of the chats list context menu for Safari (#14322)
1310f47c1d7106e49062279bcc57fc1077c0db61e278a5d1fb6212298db1be1639db4c03adb2c522ee13Akarshan Biswasakarshan@menlo.ai2025-06-29T21:07:58+05:30GitHubnoreply@github.com2025-06-29T21:07:58+05:30SYCL: disable faulty fp16 exp kernel (#14395)
1311a5d1fb6212298db1be1639db4c03adb2c522ee13a0535ffa0d35fccfec3e1a0a3bfc9dbb6054d7c0Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-29T14:38:10+02:00GitHubnoreply@github.com2025-06-29T14:38:10+02:00ggml : fix unmerged GGML_FPxx_TO_FPxx refactoring (#14443)
1312a0535ffa0d35fccfec3e1a0a3bfc9dbb6054d7c0bd9c981d7226107f18deb8344c3301450311bb8bSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-29T11:04:10+02:00GitHubnoreply@github.com2025-06-29T11:04:10+02:00ggml : implement REGLU/GEGLU/SWIGLU ops (#14158)
1313bd9c981d7226107f18deb8344c3301450311bb8b27208bf657cfe7262791df473927225e48efe482Jeff Bolzjbolz@nvidia.com2025-06-29T02:43:36-05:00GitHubnoreply@github.com2025-06-29T09:43:36+02:00vulkan: Add fusion support for RMS_NORM+MUL (#14366)
131427208bf657cfe7262791df473927225e48efe48263a7bb3c7e1c6b0a92d03b0a594d3cd501d6ed3eAman Guptaamangupta052@gmail.com2025-06-29T01:30:53+08:00GitHubnoreply@github.com2025-06-29T01:30:53+08:00CUDA: add bf16 and f32 support to cublas_mul_mat_batched (#14361)
131563a7bb3c7e1c6b0a92d03b0a594d3cd501d6ed3e00d5282c7f2a0bb05bb315fb81ca3b0f42cf9f07Jeff Bolzjbolz@nvidia.com2025-06-28T10:36:40-05:00GitHubnoreply@github.com2025-06-28T17:36:40+02:00vulkan: handle noncontig in the final case of ggml_vk_get_cpy_pipeline (#14378)
131600d5282c7f2a0bb05bb315fb81ca3b0f42cf9f07566c16fcce44876a167c37f159085afe6f84b28cJeff Bolzjbolz@nvidia.com2025-06-28T10:17:09-05:00GitHubnoreply@github.com2025-06-28T17:17:09+02:00vulkan: lock accesses of pinned_memory vector (#14333)
1317566c16fcce44876a167c37f159085afe6f84b28cb25e92774e2fa4ee3820e458d5cf43f40190f8d2Weizhao Ouyangweizhao.ouyang@arm.com2025-06-28T22:08:21+08:00GitHubnoreply@github.com2025-06-28T16:08:21+02:00model : add support for ERNIE 4.5 0.3B model (#14408)
1318b25e92774e2fa4ee3820e458d5cf43f40190f8d26609507a910aa7437aaa53fd999447de3947d998Xinpeng Dou15529241576@163.com2025-06-28T17:35:41+08:00GitHubnoreply@github.com2025-06-28T17:35:41+08:00fix async_mode bug (#14432)
13196609507a910aa7437aaa53fd999447de3947d998ceb1bf5a34d5e66e28b23dcc7a3cd83fe1e27481Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-28T09:57:07+02:00GitHubnoreply@github.com2025-06-28T09:57:07+02:00ci : fix windows build and release (#14431)
1320ceb1bf5a34d5e66e28b23dcc7a3cd83fe1e2748172babea5dea56c8a8e8420ccf731b12a5cf37854Jeff Bolzjbolz@nvidia.com2025-06-27T22:35:30-05:00GitHubnoreply@github.com2025-06-27T22:35:30-05:00vulkan: Fix GGML_VULKAN_SHADER_DEBUG_INFO (#14427)
132172babea5dea56c8a8e8420ccf731b12a5cf3785443678060c1f4cfab2f899466fd615e358677f807Georgi Gerganovggerganov@gmail.com2025-06-27T21:42:02+03:00GitHubnoreply@github.com2025-06-27T21:42:02+03:00graph : make llm_graph_context destructor virtual (#14410)
132243678060c1f4cfab2f899466fd615e358677f8078d94219a4a7f2da72ee542019ca01f36af93d1d6Georgi Gerganovggerganov@gmail.com2025-06-27T17:55:45+03:00GitHubnoreply@github.com2025-06-27T17:55:45+03:00recurrent : call balloc split_reset() in init_batch() (#14414)
13238d94219a4a7f2da72ee542019ca01f36af93d1d6f667f1e6244e1f420512fa66692b7096ff17f366Radoslav Gerganovrgerganov@gmail.com2025-06-27T16:41:40+03:00GitHubnoreply@github.com2025-06-27T16:41:40+03:00ggml : add ggml_set_rows (#14274)
1324f667f1e6244e1f420512fa66692b7096ff17f3668846aace4934ad29651ea61b8c7e3f6b0556e3d2Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-27T10:42:19+02:00GitHubnoreply@github.com2025-06-27T10:42:19+02:00convert : fix broken sentencepiece vocab (#14416)
13258846aace4934ad29651ea61b8c7e3f6b0556e3d2a01047b041aa04aeea351933658433ed004516abXuan-Son Nguyenson@huggingface.co2025-06-26T19:34:02+02:00GitHubnoreply@github.com2025-06-26T20:34:02+03:00model : gemma3n text-only (#14400)
1326a01047b041aa04aeea351933658433ed004516abb25346221dadb9101aa9dda55431dde4d3596943bandoti141645996+bandoti@users.noreply.github.com2025-06-26T13:46:53-03:00GitHubnoreply@github.com2025-06-26T13:46:53-03:00cmake: regen vulkan shaders when shaders-gen sources change (#14398)
1327b25346221dadb9101aa9dda55431dde4d3596943e8215dbb96b8fb94a24c29cdd228166fb972dbfcSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-26T15:01:14+02:00GitHubnoreply@github.com2025-06-26T15:01:14+02:00llama : return mistral-v7-tekken as default template only (#14390)
1328e8215dbb96b8fb94a24c29cdd228166fb972dbfc5783ae43599400b723b5da0569c1f848419ff3c7Georgi Gerganovggerganov@gmail.com2025-06-26T15:51:19+03:00GitHubnoreply@github.com2025-06-26T15:51:19+03:00metal : add special-case mat-vec mul for ne00 == 4 (#14385)
13295783ae43599400b723b5da0569c1f848419ff3c7bf5bcd0b857db420235e03639f0a5f218a7f8cf8Georgi Gerganovggerganov@gmail.com2025-06-26T15:50:15+03:00GitHubnoreply@github.com2025-06-26T15:50:15+03:00metal : batch rows copy in a single threadgroup (#14384)
1330bf5bcd0b857db420235e03639f0a5f218a7f8cf8716301d1b03c31875ec3b24526c48c8b1bd0fd8cAaron Teoaaron.teo1@ibm.com2025-06-26T18:41:41+08:00GitHubnoreply@github.com2025-06-26T12:41:41+02:00docs: update s390x documentation + add faq (#14389)
1331ba67d6c03fed3193987a4ef13050e6e2a4451df48f52c6f8abb2ed9ad967cc0bfb90acb073011193Yunzhe Jiayunzhe@calculet.tech2025-06-26T17:42:32+08:00Yunzhe Jiayunzhe@calculet.tech2025-06-26T17:42:32+08:00llama-calrt: infer-op: copy data from output buffer to dst-tensor.data
13328f52c6f8abb2ed9ad967cc0bfb90acb073011193d2f325130c4ed77a3dcff2c23a926587fb08e2ccYunzhe Jiayunzhe@calculet.tech2025-06-26T15:10:45+08:00Yunzhe Jiayunzhe@calculet.tech2025-06-26T15:10:45+08:00Squashed commit of the following:
1333716301d1b03c31875ec3b24526c48c8b1bd0fd8c60ef23d6c14d325d83eae5752e5de39ad268e9b0R0CKSTARyeahdongcn@gmail.com2025-06-26T12:11:59+08:00GitHubnoreply@github.com2025-06-26T12:11:59+08:00musa: enable fp16 mma (all) and cublas on qy2 (#13842)
133460ef23d6c14d325d83eae5752e5de39ad268e9b0b193d5306912a2adae0fde7481819f6ee0941bc6Aaron Teoaaron.teo1@ibm.com2025-06-26T05:49:04+08:00GitHubnoreply@github.com2025-06-25T23:49:04+02:00ggml-cpu: enable IBM NNPA Vector Intrinsics (#14317)
1335b193d5306912a2adae0fde7481819f6ee0941bc62bf9d539dd158345e3a3b096e16474af535265b4Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-25T23:26:51+02:00GitHubnoreply@github.com2025-06-25T23:26:51+02:00ggml : do not output unprintable characters on GGUF load failure (#14381)
13362bf9d539dd158345e3a3b096e16474af535265b473e53dc834c0a2336cd104473af6897197b96277Anton Mitkovanton_b_mitkov@abv.bg2025-06-25T17:09:55+01:00GitHubnoreply@github.com2025-06-25T18:09:55+02:00sycl: GGML_SYCL_DISABLE_OPT on by default for all Intel Devices (#13973)
133773e53dc834c0a2336cd104473af6897197b9627762af464227dafa1c55e0535bcb24346326748f46lhezquic_lih@quicinc.com2025-06-24T11:46:25-07:00GitHubnoreply@github.com2025-06-24T11:46:25-07:00opencl: ref count `ggml_backend_opencl_context` and refactor profiling (#14254)
133862af464227dafa1c55e0535bcb24346326748f46c148cf1946275952a79ad50b6199725f12a70411Georgi Gerganovggerganov@gmail.com2025-06-24T18:26:30+03:00GitHubnoreply@github.com2025-06-24T18:26:30+03:00batch : fix check for empty sequences in memory (#14364)
1339c148cf1946275952a79ad50b6199725f12a704111b809cee225222094a0ff5be8467240487ce4ae4Mathieu Baudiermbaudier@argeo.org2025-06-24T15:05:31+02:00GitHubnoreply@github.com2025-06-24T15:05:31+02:00cmake : use LLAMA_BUILD_NUMBER when defining LLAMA_INSTALL_VERSION (#14362)
13401b809cee225222094a0ff5be8467240487ce4ae4abf241045d09cad70dc797b0fba393ad09ee2cbeNigel Boschpnigelb@gmail.com2025-06-24T08:59:11ZGitHubnoreply@github.com2025-06-24T10:59:11+02:00server : move no API key doc to /health (#14352)
1341abf241045d09cad70dc797b0fba393ad09ee2cbe901e20bbe571fbde48d13eb188f4e7cdc7562fb6Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-24T09:31:00+02:00GitHubnoreply@github.com2025-06-24T09:31:00+02:00main : honor --verbose-prompt on interactive prompts (#14350)
1342901e20bbe571fbde48d13eb188f4e7cdc7562fb60142961a2e67909e33cdf410274b56c08c5dce7aBartowski3266127+bartowski1182@users.noreply.github.com2025-06-24T02:17:58-04:00GitHubnoreply@github.com2025-06-24T09:17:58+03:00jinja : Add Mistral-Small-3.2-24B-Instruct-2506.jinja (#14349)
13430142961a2e67909e33cdf410274b56c08c5dce7ace82bd0117bd3598300b3a089d13d401b90279c7uvosphilipp@uvos.xyz2025-06-24T01:12:56+02:00GitHubnoreply@github.com2025-06-24T01:12:56+02:00CUDA/HIP: optimize mmv paths taken for HIP devices (#14324)
1344ce82bd0117bd3598300b3a089d13d401b90279c7bf2a99e3cb06a14dce2a586450d012dc31f922aebandoti141645996+bandoti@users.noreply.github.com2025-06-23T15:30:51-03:00GitHubnoreply@github.com2025-06-23T15:30:51-03:00ci: add workflow for relocatable cmake package (#14346)
1345bf2a99e3cb06a14dce2a586450d012dc31f922ae72c6bc3f3d0cf3bf160dddf1b803fed52bcbb0a3Jeff Bolzjbolz@nvidia.com2025-06-23T08:44:48-05:00GitHubnoreply@github.com2025-06-23T15:44:48+02:00vulkan: update windows SDK in release.yml (#14344)
134672c6bc3f3d0cf3bf160dddf1b803fed52bcbb0a3defe2158dd5e250b4ef53994057a4ec03131a263Molly Sophiamollysophia379@gmail.com2025-06-23T19:56:19+08:00GitHubnoreply@github.com2025-06-23T19:56:19+08:00llama : better rwkv chat template and add missing `inputs.use_jinja` setting (#14336)
1347defe2158dd5e250b4ef53994057a4ec03131a2637b50d589a863c7631135c1226f6eab65cb406212Johannes Gäßlerjohannesg@5d6.de2025-06-23T13:11:31+02:00GitHubnoreply@github.com2025-06-23T13:11:31+02:00CUDA: mul_mat_v support for batch sizes > 1 (#14262)
13487b50d589a863c7631135c1226f6eab65cb4062123a9457df962b5883f2773f1c295e8c19df60d89fGeorgi Gerganovggerganov@gmail.com2025-06-23T12:27:35+03:00GitHubnoreply@github.com2025-06-23T12:27:35+03:00kv-cells : fix tracking of seq_pos (#14339)
13493a9457df962b5883f2773f1c295e8c19df60d89ffa4a9f2a1ccda2573189a9d4995bdf0bceb41156Jeff Bolzjbolz@nvidia.com2025-06-23T03:19:24-05:00GitHubnoreply@github.com2025-06-23T10:19:24+02:00vulkan: update windows SDK in CI (#14334)
1350fa4a9f2a1ccda2573189a9d4995bdf0bceb41156238005c2dc67426cf678baa2d54c881701693288Ed Addario29247825+EAddario@users.noreply.github.com2025-06-22T22:16:26+01:00GitHubnoreply@github.com2025-06-22T23:16:26+02:00quantize : handle user-defined pruning of whole layers (blocks) (#13037)
1351238005c2dc67426cf678baa2d54c88170169328866aba7aca9a245d71f1ddf02c7a97223529752a8Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-22T19:46:17+02:00GitHubnoreply@github.com2025-06-22T19:46:17+02:00gguf-py : fix SpecialVocab parsing when post_processor is null (#14330)
135266aba7aca9a245d71f1ddf02c7a97223529752a8f1f5e82df6222dcbaca6396c0de44df259a1694fRuikai Pengretr0@retr0.blog2025-06-23T01:28:06+08:00GitHubnoreply@github.com2025-06-23T01:28:06+08:00run : avoid double tokenization (#14327)
1353f1f5e82df6222dcbaca6396c0de44df259a1694faf3373f1adfca56119f3e4de0e6a0a8df8edf3d9Georgi Gerganovggerganov@gmail.com2025-06-22T20:10:07+03:00GitHubnoreply@github.com2025-06-22T20:10:07+03:00examples : fix is_first logic for tokenization (#14329)
1354af3373f1adfca56119f3e4de0e6a0a8df8edf3d95d5c066de8a3d2cb32f04c4d5ad1560945f30bf3uvosphilipp@uvos.xyz2025-06-22T16:51:23+02:00GitHubnoreply@github.com2025-06-22T16:51:23+02:00HIP: enable vec fattn on RDNA4 (#14323)
13555d5c066de8a3d2cb32f04c4d5ad1560945f30bf340bfa04c95c19fb42bafd4e21b5c2a7771846801yuisekiyuiseki@gmail.com2025-06-22T21:44:57+09:00GitHubnoreply@github.com2025-06-22T14:44:57+02:00mtmd : fix Pixtral OOM with large images by capping image_size to 1024 (#14326)
135640bfa04c95c19fb42bafd4e21b5c2a7771846801aa064b2eb7f7779db5e094a9d8d66d5033557de2Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-22T07:37:43+02:00GitHubnoreply@github.com2025-06-22T08:37:43+03:00common : use std::string_view now that we target c++17 (#14319)
1357aa064b2eb7f7779db5e094a9d8d66d5033557de2aa0ef5c578eef4c2adc7be1282f21bab5f3e8d26Aman Guptaamangupta052@gmail.com2025-06-22T12:39:54+08:00GitHubnoreply@github.com2025-06-22T12:39:54+08:00CUDA: add mean operation (#14313)
1358aa0ef5c578eef4c2adc7be1282f21bab5f3e8d26bb16041caef45cd4348cd6f84906b5dfec7a1f6aSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-21T18:12:05+02:00GitHubnoreply@github.com2025-06-21T18:12:05+02:00gguf-py : fix Qwen3-Embedding eos token (#14314)
1359bb16041caef45cd4348cd6f84906b5dfec7a1f6a58cba76a9aab728717509d62b67c13afd8dc227aMarkus Tavenrathmtavenrath@users.noreply.github.com2025-06-21T08:17:12+02:00GitHubnoreply@github.com2025-06-21T08:17:12+02:00Add support for VK_EXT_debug_utils to add labels to Vulkan objects. (#13792)
136058cba76a9aab728717509d62b67c13afd8dc227a67ae5312e255ae97852a4a216e2245580bfafd72Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-21T07:33:21+02:00GitHubnoreply@github.com2025-06-21T07:33:21+02:00gguf-py : fix TemplateProcessing pair when bos/eos is missing (#14312)
136167ae5312e255ae97852a4a216e2245580bfafd72692e3cdd0a069ab56411b64506a67537d767683eGeorgi Gerganovggerganov@gmail.com2025-06-21T08:04:18+03:00GitHubnoreply@github.com2025-06-21T08:04:18+03:00metal : fix thread-safety (#14300)
1362692e3cdd0a069ab56411b64506a67537d767683eb23fa0b3f40165ca3aae8ad4ee756e72f9a130ddGeorgi Gerganovggerganov@gmail.com2025-06-21T08:03:46+03:00GitHubnoreply@github.com2025-06-21T08:03:46+03:00memory : rename interface to llama_memory_context_i (#14296)
1363b23fa0b3f40165ca3aae8ad4ee756e72f9a130dd06cbedfca1587473df9b537f1dd4d6bfa2e3de13Daniel Handanielhanchen@gmail.com2025-06-20T21:32:01-07:00GitHubnoreply@github.com2025-06-21T06:32:01+02:00convert : fix Llama 4 conversion (#14311)
136406cbedfca1587473df9b537f1dd4d6bfa2e3de13b7147673f26c7ceb926a43c4734002bba291bcb8Georgi Gerganovggerganov@gmail.com2025-06-20T20:50:24+03:00Georgi Gerganovggerganov@gmail.com2025-06-20T21:02:47+03:00sync : ggml
1365b7147673f26c7ceb926a43c4734002bba291bcb8d860dd99a4178f58d1d1fa64eebc2aabc95392a7Aclyaclysia@gmail.com2025-06-18T13:34:50+02:00Georgi Gerganovggerganov@gmail.com2025-06-20T21:02:47+03:00Add `ggml_roll` (ggml/1274)
1366d860dd99a4178f58d1d1fa64eebc2aabc95392a7c959f462a0b4d42eaf930ffd72df0e435c97d5d5David Chiudavid20571015@gmail.com2025-06-21T01:43:35+08:00GitHubnoreply@github.com2025-06-20T19:43:35+02:00docs : fix the link to llama.h (#14293)
1367c959f462a0b4d42eaf930ffd72df0e435c97d5d522015b2092e291022ea3cfedc6aeb8f2643807daAman Guptaamangupta052@gmail.com2025-06-20T22:48:24+08:00GitHubnoreply@github.com2025-06-20T22:48:24+08:00CUDA: add conv_2d_transpose (#14287)
136822015b2092e291022ea3cfedc6aeb8f2643807dadd6e6d0b6a4bbe3ebfc931d1eb14db2f2b1d70afSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-20T16:37:44+02:00GitHubnoreply@github.com2025-06-20T16:37:44+02:00lint : remove trailing whitepace (#14304)
1369dd6e6d0b6a4bbe3ebfc931d1eb14db2f2b1d70af8308f98c7fb778e54bf75538f5234d8bd20915e9Ruikai Pengretr0@retr0.blog2025-06-20T22:13:06+08:00GitHubnoreply@github.com2025-06-20T07:13:06-07:00vocab : prevent tokenizer overflow (#14301)
13708308f98c7fb778e54bf75538f5234d8bd20915e96369be07359d03723f38c0a4a014ff0f698a0738Nicolò Scipionenicolo.scipione@codeplay.com2025-06-20T15:07:21+02:00GitHubnoreply@github.com2025-06-20T15:07:21+02:00sycl: add usage of enqueue_functions extension (#14244)
13716369be07359d03723f38c0a4a014ff0f698a073888fc854b4bd2e3caf10e705e6afcbbca136f0a3cChristian Kastnerckk@kvr.at2025-06-20T12:17:32ZGitHubnoreply@github.com2025-06-20T14:17:32+02:00Implement GGML_CPU_ALL_VARIANTS for PowerPC (#14286)
137288fc854b4bd2e3caf10e705e6afcbbca136f0a3ce28c1b93fd7d3f8faf9551d962e8a65fe2122e38Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-20T14:04:09+02:00GitHubnoreply@github.com2025-06-20T14:04:09+02:00llama : improve sep token handling (#14272)
1373e28c1b93fd7d3f8faf9551d962e8a65fe2122e38d27b3ca1758dfb1718e333d497ef4b68ad109bc2Diego Devesaslarengh@gmail.com2025-06-20T04:57:36-07:00GitHubnoreply@github.com2025-06-20T13:57:36+02:00cuda : synchronize graph capture and cublas handle destruction (#14288)
1374d27b3ca1758dfb1718e333d497ef4b68ad109bc29230dbe2c757e2d5071329095727d0fa9d4b85c4Georgi Gerganovggerganov@gmail.com2025-06-20T11:19:15+03:00GitHubnoreply@github.com2025-06-20T11:19:15+03:00ggml : fix repack work size for mul_mat_id (#14292)
13759230dbe2c757e2d5071329095727d0fa9d4b85c4812939a9e90f99d1bd5bb1bc6b99d12600671d50Charles Xucharles.xu@arm.com2025-06-20T09:51:01+02:00GitHubnoreply@github.com2025-06-20T10:51:01+03:00ggml: Update KleidiAI to v1.9.0 (#14277)
1376812939a9e90f99d1bd5bb1bc6b99d12600671d504c9fdfbe1580a66fd7d77c77418ce2c606a29fddGeorgi Gerganovggerganov@gmail.com2025-06-20T10:50:27+03:00GitHubnoreply@github.com2025-06-20T10:50:27+03:00model : more uniform output id handling (#14275)
13774c9fdfbe1580a66fd7d77c77418ce2c606a29fdd9eaa51e7f08593f123f00136591179a8f5956ecdGeorgi Gerganovggerganov@gmail.com2025-06-20T10:14:14+03:00GitHubnoreply@github.com2025-06-20T10:14:14+03:00ubatch : new splitting logic (#14217)
1378d2f325130c4ed77a3dcff2c23a926587fb08e2cc6391ee7ffd97e723dfbdcc32f001e1b95a63cba6Yunzhe Jiayunzhe@calculet.tech2025-06-17T11:53:33+08:00Yunzhe Jiayunzhe@calculet.tech2025-06-20T09:57:18+08:00calrt: add input_token in graph; input_token will be copy to calrt_in_buffer
13799eaa51e7f08593f123f00136591179a8f5956ecd8f71d0f3e86ccbba059350058af8758cafed73e6Aman Guptaamangupta052@gmail.com2025-06-20T09:50:24+08:00GitHubnoreply@github.com2025-06-20T09:50:24+08:00CUDA: add conv_2d_dw (#14265)
13808f71d0f3e86ccbba059350058af8758cafed73e6381174bbdaf10d6a80dc2099f284b20544d86962Diego Devesaslarengh@gmail.com2025-06-19T12:24:14-07:00GitHubnoreply@github.com2025-06-19T21:24:14+02:00ggml-cpu : remove unnecesary arm feature detection (#14281)
1381381174bbdaf10d6a80dc2099f284b20544d86962d67341dc18fc5cc63362880ab2f8f9ecfc7932e7Alex Trotta44127594+Ahajha@users.noreply.github.com2025-06-19T09:56:12-04:00GitHubnoreply@github.com2025-06-19T15:56:12+02:00gguf-py : make sentencepiece optional (#14200)
1382d67341dc18fc5cc63362880ab2f8f9ecfc7932e7456af35eb70177b8dd5779b6d4c21bb020f9cebdaa956aa956@users.noreply.github.com2025-06-19T16:01:03+03:00GitHubnoreply@github.com2025-06-19T16:01:03+03:00server : add server parameters for draft model cache type (#13782)
1383456af35eb70177b8dd5779b6d4c21bb020f9cebd600e3e9b50c1f0c9fc4a70356241fd87f00e8e14fanyangfanyang89@outlook.com2025-06-19T20:49:48+08:00GitHubnoreply@github.com2025-06-19T14:49:48+02:00build : suppress gcc15 compile warnings (#14261)
1384600e3e9b50c1f0c9fc4a70356241fd87f00e8e14fffcce535ebbdc25f81966a15b758658788e7466Anton Mitkovanton_b_mitkov@abv.bg2025-06-19T11:40:21+01:00GitHubnoreply@github.com2025-06-19T11:40:21+01:00sycl: Cleanup codepaths in Get Rows in sycl backend (#14215)
1385fffcce535ebbdc25f81966a15b758658788e74665fc7856815920f828f9e90cb759ac82c7f0c1ea5bashayer hijjibashayer.hijji@gmail.com2025-06-19T13:24:12+03:00GitHubnoreply@github.com2025-06-19T12:24:12+02:00llama-bench : add --no-warmup flag (#14224) (#14270)
13865fc7856815920f828f9e90cb759ac82c7f0c1ea5faed5a5f5dde4d816adecdccb4f4682a04126f92pqnet119850+pqnet@users.noreply.github.com2025-06-19T12:21:40+02:00GitHubnoreply@github.com2025-06-19T12:21:40+02:00convert : fix remote option in Windows (#14100)
1387faed5a5f5dde4d816adecdccb4f4682a04126f9210bb545c5b54175ed9874ad8d187effa2bcb4b5fAaron Teoaaron.teo1@ibm.com2025-06-19T17:48:54+08:00GitHubnoreply@github.com2025-06-19T11:48:54+02:00llamafile : support s390x SIMD instruction set (#14273)
138810bb545c5b54175ed9874ad8d187effa2bcb4b5fedc4a29effe716956fdfd2bc5b9cba2a6d8492f80cc4mpicard12@live.de2025-06-19T09:15:42+02:00GitHubnoreply@github.com2025-06-19T09:15:42+02:00Vulkan: Set device max size for host memory to avoid OOM warning and fallback to CPU buffer (#14249)
1389edc4a29effe716956fdfd2bc5b9cba2a6d8492f8ed3290ab34493fd3d2e0f925f816a401da4f2dfdGabe Goodhartgabe.l.hart@gmail.com2025-06-19T00:08:14-05:00GitHubnoreply@github.com2025-06-19T08:08:14+03:00memory : Hybrid recurrent cache (#13979)
1390ed3290ab34493fd3d2e0f925f816a401da4f2dfd8d947136546773f6410756f37fcc5d3e65b8135dGeorgi Gerganovggerganov@gmail.com2025-06-19T08:05:21+03:00GitHubnoreply@github.com2025-06-19T08:05:21+03:00metal : add mean kernel (#14267)
13918d947136546773f6410756f37fcc5d3e65b8135d50d2227953ca9024f04255b4f116d06fcc0db74cAaron Teoaaron.teo1@ibm.com2025-06-19T01:10:26+08:00GitHubnoreply@github.com2025-06-18T18:10:26+01:00docs: add s390x build documentation (#14264)
139250d2227953ca9024f04255b4f116d06fcc0db74c6231c5cd6d49d61511f328b5f43322407df90a91Aaron Teoaaron.teo1@ibm.com2025-06-19T01:10:08+08:00GitHubnoreply@github.com2025-06-18T18:10:08+01:00ggml-cpu: reduce asm calls for hsum (#14037)
13936231c5cd6d49d61511f328b5f43322407df90a91ef035803eb9dbc306ea9a8ff82e30af12b567cf7Aaron Teoaaron.teo1@ibm.com2025-06-19T01:06:49+08:00GitHubnoreply@github.com2025-06-18T18:06:49+01:00ggml-cpu: fix uncaught underscore terminators (#14023)
1394ef035803eb9dbc306ea9a8ff82e30af12b567cf7413977de32e90712ecec84d0b9c738847da8dc02Charles Xucharles.xu@arm.com2025-06-18T13:40:07+02:00GitHubnoreply@github.com2025-06-18T12:40:07+01:00ggml: Add Apple support for GGML_CPU_ALL_VARIANTS (#14258)
1395413977de32e90712ecec84d0b9c738847da8dc0295402553a5effc61ddc9e29c7bcb56f71311dd4aXuan-Son Nguyenson@huggingface.co2025-06-18T10:43:57+02:00GitHubnoreply@github.com2025-06-18T10:43:57+02:00mtmd : refactor llava-uhd preprocessing logic (#14247)
139695402553a5effc61ddc9e29c7bcb56f71311dd4a3865cff4f5b84c16119590efd6ce537789a27715Xuan-Son Nguyenson@huggingface.co2025-06-18T09:58:43+02:00GitHubnoreply@github.com2025-06-18T09:58:43+02:00llama-chat : fix multiple system message for gemma, orion (#14246)
13973865cff4f5b84c16119590efd6ce537789a27715d03172cc797b6adbbc00bfc09caf614fb0f895a0Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-18T09:52:07+02:00GitHubnoreply@github.com2025-06-18T09:52:07+02:00convert : fix null head_dim AutoConfig regression (#14248)
1398d03172cc797b6adbbc00bfc09caf614fb0f895a0dd8e59f4435342eda93c5b0cf4109e21c9c7d0ebGeorgi Gerganovggerganov@gmail.com2025-06-18T09:58:23+03:00Georgi Gerganovggerganov@gmail.com2025-06-18T09:59:21+03:00sync : ggml
1399dd8e59f4435342eda93c5b0cf4109e21c9c7d0ebbbe98d27840453c8787d18470963530fdc27d89fDaniel Beveniusdaniel.bevenius@gmail.com2025-06-13T15:06:42+02:00Georgi Gerganovggerganov@gmail.com2025-06-18T09:59:21+03:00ggml : disable warnings for tests when using MSVC (ggml/1273)
1400bbe98d27840453c8787d18470963530fdc27d89fc2056ed6d461e6d5432f04f221e221ab795dc652Daniel Beveniusdaniel.bevenius@gmail.com2025-06-13T09:05:44+02:00Georgi Gerganovggerganov@gmail.com2025-06-18T09:59:21+03:00ggml : remove unused ggml_context_container (ggml/1272)
1401c2056ed6d461e6d5432f04f221e221ab795dc652c46503014db0d63fa7b1b28c58adfb51054e2decDaniel Beveniusdaniel.bevenius@gmail.com2025-06-12T12:27:09+02:00Georgi Gerganovggerganov@gmail.com2025-06-18T09:59:21+03:00examples : include examples in msvc disable warn (ggml/1270)
1402c46503014db0d63fa7b1b28c58adfb51054e2dec860a9e4eeff3eb2e7bd1cc38f65787cc6c8177afbandoti141645996+bandoti@users.noreply.github.com2025-06-17T17:33:25-03:00GitHubnoreply@github.com2025-06-17T22:33:25+02:00cmake: remove shader-gen step-targets from ggml-vulkan (#14226)
1403860a9e4eeff3eb2e7bd1cc38f65787cc6c8177affe9d60e74a6cb71bcaed2029377bfa2872b4abb0xctanxc-tan@outlook.com2025-06-17T17:58:32+08:00GitHubnoreply@github.com2025-06-17T12:58:32+03:00ggml-cpu : remove the weak alias trick (#14221)
1404fe9d60e74a6cb71bcaed2029377bfa2872b4abb0e434e69183fd9e1031f4445002083178c331a28bR0CKSTARyeahdongcn@gmail.com2025-06-17T17:48:08+08:00GitHubnoreply@github.com2025-06-17T17:48:08+08:00musa: fix build warning (unused variable) (#14231)
1405e434e69183fd9e1031f4445002083178c331a28b89fea80d298184d1cd93564f48e060d9f541f4b4Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-16T21:58:42+02:00GitHubnoreply@github.com2025-06-16T21:58:42+02:00common : suggest --jinja when autodetection fails (#14222)
140689fea80d298184d1cd93564f48e060d9f541f4b46adc3c3ebc029af058ac950a8e2a825fdf18ecc6Georgi Gerganovggerganov@gmail.com2025-06-16T22:33:27+03:00GitHubnoreply@github.com2025-06-16T22:33:27+03:00server : fix incorrect usage of llama_get_embeddings() (#14225)
14076adc3c3ebc029af058ac950a8e2a825fdf18ecc60dbcabde8c006d5cf781ca0fe071c41559572a72Diego Devesaslarengh@gmail.com2025-06-16T08:11:43-07:00GitHubnoreply@github.com2025-06-16T08:11:43-07:00llama : add thread safety test (#14035)
14080dbcabde8c006d5cf781ca0fe071c41559572a72ad590be98c83217fcf1a101d78d9ab389fd5dc0bbandoti141645996+bandoti@users.noreply.github.com2025-06-16T10:32:13-03:00GitHubnoreply@github.com2025-06-16T10:32:13-03:00cmake: clean up external project logic for vulkan-shaders-gen (#14179)
1409ad590be98c83217fcf1a101d78d9ab389fd5dc0b7d6d91babfa129906b39c9099eca4234c44f4f1eĐinh Trọng Huy77562200+huydt84@users.noreply.github.com2025-06-16T21:53:41+09:00GitHubnoreply@github.com2025-06-16T14:53:41+02:00model : add NeoBERT (#14164)
14107d6d91babfa129906b39c9099eca4234c44f4f1ed3e64b9f490cee41b7b9aa275dae2f6568ae3054uvosphilipp@uvos.xyz2025-06-16T13:47:38+02:00GitHubnoreply@github.com2025-06-16T13:47:38+02:00HIP: disable rocwmma on gfx12 by default until rocm 7.0 (#14202)
1411d3e64b9f490cee41b7b9aa275dae2f6568ae30543ba0d843c6bd3faea5cf5e53dc7f3c82be20bffbGeorgi Gerganovggerganov@gmail.com2025-06-16T14:14:00+03:00GitHubnoreply@github.com2025-06-16T14:14:00+03:00llama : rework embeddings logic (#14208)
14123ba0d843c6bd3faea5cf5e53dc7f3c82be20bffb0bf49eb668bb95b50e41583e22aaf60ddade1fbeCharles Xucharles.xu@arm.com2025-06-16T11:47:57+02:00GitHubnoreply@github.com2025-06-16T11:47:57+02:00ggml: Add Android support for GGML_CPU_ALL_VARIANTS (#14206)
14130bf49eb668bb95b50e41583e22aaf60ddade1fbe4ad243677bca6c97f14dbc187b2116b51fcb7ffdBartowski3266127+bartowski1182@users.noreply.github.com2025-06-16T09:16:06+01:00GitHubnoreply@github.com2025-06-16T10:16:06+02:00convert : remove arcee change in convert_hf_to_gguf_update.py (#14207)
14144ad243677bca6c97f14dbc187b2116b51fcb7ffdc89c2d1ab94b11845240b7d3313c87691ea18d88Đinh Trọng Huy77562200+huydt84@users.noreply.github.com2025-06-16T16:20:59+09:00GitHubnoreply@github.com2025-06-16T09:20:59+02:00gguf-py : allow key override when adding value to GGUFWriter (#14194)
1415c89c2d1ab94b11845240b7d3313c87691ea18d883555b3004ba7687be3d734acade52a3345758aa4Jeff Bolzjbolz@nvidia.com2025-06-16T00:21:08-06:00GitHubnoreply@github.com2025-06-16T08:21:08+02:00vulkan: mutex around vkQueueSubmit (#14127)
14163555b3004ba7687be3d734acade52a3345758aa4d7da8dc83a03b30e1ec10317080082ea76840c38xctanxc-tan@outlook.com2025-06-16T13:54:15+08:00GitHubnoreply@github.com2025-06-16T13:54:15+08:00ggml-cpu : rework weak alias on apple targets (#14146)
1417d7da8dc83a03b30e1ec10317080082ea76840c38cd355eda7df1898d25d433b4bdaa4b4b479e0badBartowski3266127+bartowski1182@users.noreply.github.com2025-06-16T00:04:06+01:00GitHubnoreply@github.com2025-06-16T01:04:06+02:00model : Add support for Arcee AI's upcoming AFM model (#14185)
1418cd355eda7df1898d25d433b4bdaa4b4b479e0bad30e5b01de2a0bcddc7c063c8ef0802703a958417Eric Curtinecurtin@redhat.com2025-06-15T23:36:22+02:00GitHubnoreply@github.com2025-06-15T23:36:22+02:00server : When listening on a unix domain socket don't print http:// and port (#14180)
141930e5b01de2a0bcddc7c063c8ef0802703a958417e54b394082de242be4ee2e692b11fcc8d4eba371Ed Addario29247825+EAddario@users.noreply.github.com2025-06-15T17:53:45+01:00GitHubnoreply@github.com2025-06-15T18:53:45+02:00quantize : change int to unsigned int for KV overrides (#14197)
1420e54b394082de242be4ee2e692b11fcc8d4eba3712c2caa444341d99c87ff153f142c2d4762a776a2uvosphilipp@uvos.xyz2025-06-15T17:30:13+02:00GitHubnoreply@github.com2025-06-15T17:30:13+02:00CUDA/HIP: fix ssm_scan on devices where warp size is not 32 (#14196)
14212c2caa444341d99c87ff153f142c2d4762a776a25fce5f948df8f189a5401a8ecaa9753106e75abbuvosphilipp@uvos.xyz2025-06-15T15:45:27+02:00GitHubnoreply@github.com2025-06-15T15:45:27+02:00HIP: Replace usage of depricated preprocessor macro __AMDGCN_WAVEFRONT_SIZE__ (#14183)
14225fce5f948df8f189a5401a8ecaa9753106e75abb9ae4143bc6ecb4c2f0f0301578f619f6c201b857Georgi Gerganovggerganov@gmail.com2025-06-15T10:52:11+03:00GitHubnoreply@github.com2025-06-15T10:52:11+03:00kv-cache : fix use-after-move of defrag info (#14189)
14239ae4143bc6ecb4c2f0f0301578f619f6c201b857c311ac664d68d10781a3e7b9f02d9d9520837d80Mikko Juolamikjuo@gmail.com2025-06-15T00:52:06-07:00GitHubnoreply@github.com2025-06-15T09:52:06+02:00model : add dots.llm1 architecture support (#14044) (#14118)
1424c311ac664d68d10781a3e7b9f02d9d9520837d80b9912ac570de8945ae9383c9ca8291027bf287ddGeorgi Gerganovggerganov@gmail.com2025-06-15T10:08:58+03:00GitHubnoreply@github.com2025-06-15T10:08:58+03:00cparams : rename LLAMA_MAX_PARALLEL_SEQUENCES to LLAMA_MAX_SEQ (#14188)
1425b9912ac570de8945ae9383c9ca8291027bf287dd00ba7726100d7e1941d9f5a06f56a7559945b33cGeorgi Gerganovggerganov@gmail.com2025-06-15T09:18:37+03:00GitHubnoreply@github.com2025-06-15T09:18:37+03:00batch : auto-gen positions + verify multi-sequence input (#14177)
142600ba7726100d7e1941d9f5a06f56a7559945b33c3cb203c89f60483e349f841684173446ed23c28fPepijn de Vosme@pepijndevos.nl2025-06-15T08:06:37+02:00GitHubnoreply@github.com2025-06-15T08:06:37+02:00docs : remove WIP since PR has been merged (#13912)
14273cb203c89f60483e349f841684173446ed23c28f2e42be42bd6bf1dcc643d6ac4e77419bfe5dd24fPiotrpiotr.stankiewicz@docker.com2025-06-14T18:25:15+02:00GitHubnoreply@github.com2025-06-14T17:25:15+01:00llama-chat : Do not throw when tool parsing fails (#14012)
14282e42be42bd6bf1dcc643d6ac4e77419bfe5dd24ffb85a288d72abbd5e5daa8de96e6f8bfa7b5ab46Aman Guptaamangupta052@gmail.com2025-06-14T16:34:20+08:00GitHubnoreply@github.com2025-06-14T10:34:20+02:00compare-llama-bench: add option to plot (#14169)
1429fb85a288d72abbd5e5daa8de96e6f8bfa7b5ab4640643edb86eb10b471b0f57d4f3f7eb0e06a0df7Georgi Gerganovggerganov@gmail.com2025-06-13T20:03:05+03:00GitHubnoreply@github.com2025-06-13T20:03:05+03:00vocab : fix build (#14175)
143040643edb86eb10b471b0f57d4f3f7eb0e06a0df73cfbbdb44e08fd19429fed6cc85b982a91f0efd5Svetlozar Georgiev55534064+sgeor255@users.noreply.github.com2025-06-13T17:32:56+01:00GitHubnoreply@github.com2025-06-13T18:32:56+02:00sycl: fix docker image (#14144)
14313cfbbdb44e08fd19429fed6cc85b982a91f0efd580709b70a2f87c13ccaf1480b799393109996789Guy Goldenbergguy110698@gmail.com2025-06-13T19:20:25+03:00GitHubnoreply@github.com2025-06-13T19:20:25+03:00Merge commit from fork
143280709b70a2f87c13ccaf1480b79939310999678926ff3685bfbaa4c8838d7afd988b17dd5eb99f92Georgi Gerganovggerganov@gmail.com2025-06-13T18:35:00+03:00GitHubnoreply@github.com2025-06-13T18:35:00+03:00batch : add LLAMA_BATCH_DEBUG environment variable (#14172)
143326ff3685bfbaa4c8838d7afd988b17dd5eb99f9260c666347becacff81cd4bc9a52038ba71038e41ddpasa112642920+ddpasa@users.noreply.github.com2025-06-13T15:17:53+02:00GitHubnoreply@github.com2025-06-13T15:17:53+02:00docs : Update multimodal.md (#14122)
143460c666347becacff81cd4bc9a52038ba71038e41b7cc7745e38141060845145af0a1fd489d8e3e33Georgi Gerganovggerganov@gmail.com2025-06-13T13:47:55+03:00GitHubnoreply@github.com2025-06-13T13:47:55+03:00batch : rework llama_batch_allocr (#14153)
1435b7cc7745e38141060845145af0a1fd489d8e3e33cc8d08187918c6f643c3ffabb7b1ac21aa19f3d1Georgi Gerganovggerganov@gmail.com2025-06-13T11:55:44+03:00GitHubnoreply@github.com2025-06-13T11:55:44+03:00readme : remove survey link (#14168)
1436cc8d08187918c6f643c3ffabb7b1ac21aa19f3d1d714dadb57d8feaa03d13b79345a4c3382172d61Christian Kastnerckk@kvr.at2025-06-13T08:38:52ZGitHubnoreply@github.com2025-06-13T10:38:52+02:00cmake: Add ability to pass in LLAMA_BUILD_NUMBER/COMMIT (#14167)
1437d714dadb57d8feaa03d13b79345a4c3382172d61ffad04397399ea1650fda6560c7c753059804876Đinh Trọng Huy77562200+huydt84@users.noreply.github.com2025-06-13T17:34:08+09:00GitHubnoreply@github.com2025-06-13T11:34:08+03:00pooling : make cls_b and cls_out_b optional (#14165)
1438ffad04397399ea1650fda6560c7c7530598048760889eba570126f8a2f5a0e88fde776bbc91cca66Georgi Gerganovggerganov@gmail.com2025-06-13T11:18:25+03:00GitHubnoreply@github.com2025-06-13T11:18:25+03:00server : fix SWA condition for full context reprocess (#14163)
14390889eba570126f8a2f5a0e88fde776bbc91cca66c61285e7396c8e526fe7794c19e8d4f1c99bfc51Anton Mitkovanton.mitkov@codeplay.com2025-06-13T08:51:39+01:00GitHubnoreply@github.com2025-06-13T08:51:39+01:00sycl: Adding additional cpy dbg print output (#14034)
1440c61285e7396c8e526fe7794c19e8d4f1c99bfc5109cf2c7c655c90e53e100f29b830a788bab0653dEwan Crawfordewan@codeplay.com2025-06-13T08:45:37+01:00GitHubnoreply@github.com2025-06-13T08:45:37+01:00SYCL: Bump oneMath commit (#14152)
144109cf2c7c655c90e53e100f29b830a788bab0653dc33fe8b8c4427202706b1434e9fc8ab5752c9cacChristian Kastnerckk@kvr.at2025-06-13T06:51:34ZGitHubnoreply@github.com2025-06-13T09:51:34+03:00cmake : Improve build-info.cpp generation (#14156)
1442c33fe8b8c4427202706b1434e9fc8ab5752c9caced52f3668e633423054a4eab61bb7efee47025abGeorgi Gerganovggerganov@gmail.com2025-06-13T08:03:54+03:00GitHubnoreply@github.com2025-06-13T08:03:54+03:00vocab : prevent heap overflow when vocab is too small (#14145)
14436391ee7ffd97e723dfbdcc32f001e1b95a63cba6b9850050a5b5147cb2e6720b98ed8c1e8e02eff9Yunzhe Jiayunzhe@calculet.tech2025-06-13T10:58:06+08:00Yunzhe Jiayunzhe@calculet.tech2025-06-13T10:58:06+08:00update .gitignore to update cal_test.gguf
1444b9850050a5b5147cb2e6720b98ed8c1e8e02eff9933b205c32551848589b8339437d778cb4f2a46dYunzhe Jiayunzhe@calculet.tech2025-06-06T17:25:52+08:00Yunzhe Jiayunzhe@calculet.tech2025-06-13T09:20:23+08:00calrt: add call chain for using calrt::infer
1445ed52f3668e633423054a4eab61bb7efee47025aba681b4ba83a61dce71a4f24e558efe7278d8b1a9Anton Mitkovanton.mitkov@codeplay.com2025-06-12T14:15:11+01:00GitHubnoreply@github.com2025-06-12T15:15:11+02:00sycl: Remove not needed copy f16->f32 for dnnl mul mat (#14125)
1446a681b4ba83a61dce71a4f24e558efe7278d8b1a97d516443dd7766569110b38e6374649bee6eb1c4Georgi Gerganovggerganov@gmail.com2025-06-12T14:43:09+03:00GitHubnoreply@github.com2025-06-12T14:43:09+03:00readme : remove project status link (#14149)
1447933b205c32551848589b8339437d778cb4f2a46d33fea9a8592fd338abe279b89350a3af2e0ff1f7Yunzhe Jiayunzhe@calculet.tech2025-06-12T17:47:19+08:00Yunzhe Jiayunzhe@calculet.tech2025-06-12T17:50:26+08:00gguf-py: add cal_test.py
14487d516443dd7766569110b38e6374649bee6eb1c4f6e1a7aa8787b5c00acba6370cb70a0beff48b1eGeorgi Gerganovggerganov@gmail.com2025-06-12T11:51:38+03:00GitHubnoreply@github.com2025-06-12T11:51:38+03:00server : re-enable SWA speculative decoding (#14131)
1449f6e1a7aa8787b5c00acba6370cb70a0beff48b1ec3ee46fab49a765d2e32e171e9ed7a5fa121dd9cGeorgi Gerganovggerganov@gmail.com2025-06-12T11:50:01+03:00GitHubnoreply@github.com2025-06-12T11:50:01+03:00context : simplify output counting logic during decode (#14142)
1450c3ee46fab49a765d2e32e171e9ed7a5fa121dd9ce2c0b6e46a5596665569ae765f0993cea2619af6Georgi Gerganovggerganov@gmail.com2025-06-12T11:49:26+03:00GitHubnoreply@github.com2025-06-12T11:49:26+03:00batch : remove logits_all flag (#14141)
1451e2c0b6e46a5596665569ae765f0993cea2619af69596506965f65be5d802ecef6a315fe43d2391a8Georgi Gerganovggerganov@gmail.com2025-06-12T10:14:24+03:00GitHubnoreply@github.com2025-06-12T10:14:24+03:00cmake : handle whitepsaces in path during metal build (#14126)
14529596506965f65be5d802ecef6a315fe43d2391a8a20b2b05bce6622c585459ebf46f142f113d021cGeorgi Gerganovggerganov@gmail.com2025-06-12T10:02:15+03:00GitHubnoreply@github.com2025-06-12T10:02:15+03:00kv-cache : fix split_equal handling in unified implementation (#14130)
1453a20b2b05bce6622c585459ebf46f142f113d021c2e89f76b7af2c0b827be785e445f2e2b3e52e1cacompiladegit@compilade.net2025-06-12T02:56:04-04:00GitHubnoreply@github.com2025-06-12T02:56:04-04:00context : round n_tokens to next multiple of n_seqs when reserving (#14140)
14542e89f76b7af2c0b827be785e445f2e2b3e52e1ca532802f938c6a18cc6a704057ab571f253fd77edbandoti141645996+bandoti@users.noreply.github.com2025-06-11T17:19:44-03:00GitHubnoreply@github.com2025-06-11T17:19:44-03:00common: fix issue with regex_escape routine on windows (#14133)
1455532802f938c6a18cc6a704057ab571f253fd77edd4e0d95cf581f50c9a21d06eaecae2dd580076bdChristian Kastnerckk@kvr.at2025-06-11T19:07:44ZGitHubnoreply@github.com2025-06-11T21:07:44+02:00Implement GGML_CPU_ALL_VARIANTS for ARM (#14080)
1456d4e0d95cf581f50c9a21d06eaecae2dd580076bdcc66a7f78f95dcb5208420f4dd1abc3fc6aec0ccSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-11T19:04:23+02:00GitHubnoreply@github.com2025-06-11T19:04:23+02:00chore : clean up relative source dir paths (#14128)
1457cc66a7f78f95dcb5208420f4dd1abc3fc6aec0ccbd248d4dc7265437e1918dc53ae3f49a0c592e5fSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-11T17:16:32+02:00GitHubnoreply@github.com2025-06-11T17:16:32+02:00tests : add test-tokenizers-repo (#14017)
1458bd248d4dc7265437e1918dc53ae3f49a0c592e5f7781e5fe99f2a4fc1c8af0a8488eedac4644cb72Jeff Bolzjbolz@nvidia.com2025-06-11T09:48:52-05:00GitHubnoreply@github.com2025-06-11T09:48:52-05:00vulkan: Better thread-safety for command pools/buffers (#14116)
14597781e5fe99f2a4fc1c8af0a8488eedac4644cb7289a184fa7125b7c3e2fb567337cc2bd7bc2d376cAmanamangupta052@gmail.com2025-06-11T22:42:25+08:00GitHubnoreply@github.com2025-06-11T16:42:25+02:00webui: Wrap long numbers instead of infinite horizontal scroll (#14062)
146089a184fa7125b7c3e2fb567337cc2bd7bc2d376c2baf07727f921d9a4a1b63a2eff941e95d0488edGeorgi Gerganovggerganov@gmail.com2025-06-11T16:48:45+03:00GitHubnoreply@github.com2025-06-11T16:48:45+03:00kv-cache : relax SWA masking condition (#14119)
14612baf07727f921d9a4a1b63a2eff941e95d0488ed7ae2932116a4de3141cbc8c488f7f6e4e06d8171Taylorquantumtraveling@gmail.com2025-06-11T06:43:43-04:00GitHubnoreply@github.com2025-06-11T13:43:43+03:00server : pass default --keep argument (#14120)
14627ae2932116a4de3141cbc8c488f7f6e4e06d81711f7d50b2936023b26eb218e944e62834b80a2ce0Georgi Gerganovggerganov@gmail.com2025-06-11T12:52:45+03:00GitHubnoreply@github.com2025-06-11T12:52:45+03:00kv-cache : add LLAMA_KV_CACHE_DEBUG environment variable (#14121)
14631f7d50b2936023b26eb218e944e62834b80a2ce04c763c8d1b4d4de20bf364ec1837430783cba984Jeff Bolzjbolz@nvidia.com2025-06-11T00:19:25-05:00GitHubnoreply@github.com2025-06-11T07:19:25+02:00vulkan: Track descriptor pools/sets per-context (#14109)
14644c763c8d1b4d4de20bf364ec1837430783cba984dad5c44398b78467943ed0a603ea427fe9f6fc62lhezquic_lih@quicinc.com2025-06-10T16:55:58-07:00GitHubnoreply@github.com2025-06-10T16:55:58-07:00opencl: add `mul_mv_id_q4_0_f32_8x_flat` (#14003)
1465dad5c44398b78467943ed0a603ea427fe9f6fc6255f6b9fa6563f6ae49113f9abdc980c12348cc1ccompiladegit@compilade.net2025-06-10T18:20:14-04:00GitHubnoreply@github.com2025-06-10T18:20:14-04:00kv-cache : avoid modifying recurrent cells when setting inputs (#13834)
146655f6b9fa6563f6ae49113f9abdc980c12348cc1c3678b838bb71eaccbaeb479ff38c2e12bfd2f960Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-10T23:29:52+02:00GitHubnoreply@github.com2025-06-10T23:29:52+02:00convert : fix duplicate key DeepSeek-R1 conversion error (#14103)
14673678b838bb71eaccbaeb479ff38c2e12bfd2f960652b70e6678d503626f3c9d33831ba604b473401Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-10T18:02:08+02:00GitHubnoreply@github.com2025-06-10T18:02:08+02:00llama : support GEGLU for jina-bert-v2 (#14090)
1468652b70e6678d503626f3c9d33831ba604b4734013a12db23b6918682ccb70ab15d897f11fe5fc320Jeff Bolzjbolz@nvidia.com2025-06-10T10:53:47-05:00GitHubnoreply@github.com2025-06-10T10:53:47-05:00vulkan: force device 0 in CI (#14106)
14693a12db23b6918682ccb70ab15d897f11fe5fc320ae92c1855b1a5b604fa1bfcced19a556ab3e78c5Juk Armstrong69222624+jukofyork@users.noreply.github.com2025-06-10T16:48:07+01:00GitHubnoreply@github.com2025-06-10T16:48:07+01:00Fixed spec timings to: accepted/tested instead of accepted/drafted (#14104)
1470ae92c1855b1a5b604fa1bfcced19a556ab3e78c5b7ce1ad1e332da5909772d173a7e6748fbd1887aGeorgi Gerganovggerganov@gmail.com2025-06-10T17:37:45+03:00Georgi Gerganovggerganov@gmail.com2025-06-10T18:39:33+03:00sync : ggml
1471b7ce1ad1e332da5909772d173a7e6748fbd1887a97340b4c9924be86704dbf155e97c8319849ee19Georgi Gerganovggerganov@gmail.com2025-06-10T11:34:10+03:00Georgi Gerganovggerganov@gmail.com2025-06-10T18:39:33+03:00ggml : fix weak alias win32 (whisper/0)
147297340b4c9924be86704dbf155e97c8319849ee192bb0467043258bdc58dbaefb33786f1731b389370cc4mpicard12@live.de2025-06-10T14:01:33+02:00GitHubnoreply@github.com2025-06-10T13:01:33+01:00Vulkan: Don't default to CPU device (like llvmpipe), even if no other device is available, to allow fallback to CPU backend (#14099)
14732bb0467043258bdc58dbaefb33786f1731b38937b8e2194efc529378be45ab9b27d6648a5b81458aIsaac McFadyenisaac@imcf.me2025-06-10T02:41:01-04:00GitHubnoreply@github.com2025-06-10T09:41:01+03:00rpc : nicer error messages for RPC server crash (#14076)
1474b8e2194efc529378be45ab9b27d6648a5b81458a1a3b5e80f77c51e6d9fb7762f19b9e61fceca3deGeorgi Gerganovggerganov@gmail.com2025-06-10T09:20:51+03:00Georgi Gerganovggerganov@gmail.com2025-06-10T09:21:56+03:00sync : ggml
14751a3b5e80f77c51e6d9fb7762f19b9e61fceca3de1f63e75f3b5dc7f44dbe63c8a41d23958fe95bc0Kai Pastordg0yt@darc.de2025-06-03T12:33:28+02:00Georgi Gerganovggerganov@gmail.com2025-06-10T09:21:56+03:00Add in-build ggml::ggml ALIAS library (ggml/1260)
14761f63e75f3b5dc7f44dbe63c8a41d23958fe95bc040cbf571c9210031340f022d104317e89a1a6bb2Georgi Gerganovggerganov@gmail.com2025-06-09T23:05:02+03:00GitHubnoreply@github.com2025-06-09T23:05:02+03:00metal : use less stack memory in FA kernel (#14088)
147740cbf571c9210031340f022d104317e89a1a6bb27f4fbe5183b23b6b2e25fd1ccc5d1fa8bb010cb7Georgi Gerganovggerganov@gmail.com2025-06-09T23:04:35+03:00GitHubnoreply@github.com2025-06-09T23:04:35+03:00kv-cache : fix shift and defrag logic (#14081)
14787f4fbe5183b23b6b2e25fd1ccc5d1fa8bb010cb7f470bc36bed4d836b9de5a483fa0dfaee176d6f5Diego Devesaslarengh@gmail.com2025-06-09T11:03:09-07:00GitHubnoreply@github.com2025-06-09T20:03:09+02:00llama : allow building all tests on windows when not using shared libs (#13980)
1479f470bc36bed4d836b9de5a483fa0dfaee176d6f58f47e25f56e9792093b7497c68e9f80bab82ed19xctanaxunlei@gmail.com2025-06-09T22:47:13+08:00GitHubnoreply@github.com2025-06-09T16:47:13+02:00ggml-cpu : split arch-specific implementations (#13892)
14808f47e25f56e9792093b7497c68e9f80bab82ed19201b31dc2e6fef3de598280b53fd3b69420a4e49Diego Devesaslarengh@gmail.com2025-06-09T07:36:26-07:00GitHubnoreply@github.com2025-06-09T16:36:26+02:00cuda : fix device sync on buffer clear (#14033)
1481201b31dc2e6fef3de598280b53fd3b69420a4e49e21d2d4ae26d6c5daf1602a1061aa4f8e722ca57Georgi Gerganovggerganov@gmail.com2025-06-09T17:17:31+03:00GitHubnoreply@github.com2025-06-09T17:17:31+03:00graph : fix geglu (#14077)
1482e21d2d4ae26d6c5daf1602a1061aa4f8e722ca57dc0623fddb661926e0998538895155e4f081ff09Xinpeng Dou15529241576@163.com2025-06-09T19:47:39+08:00GitHubnoreply@github.com2025-06-09T19:47:39+08:00CANN: Simplify the environment variable setting(#13104)
1483dc0623fddb661926e0998538895155e4f081ff0987d34b381d5868e75586210ec17b5ef5deddc276R0CKSTARyeahdongcn@gmail.com2025-06-09T18:01:17+08:00GitHubnoreply@github.com2025-06-09T12:01:17+02:00webui: fix sidebar being covered by main content (#14082)
148487d34b381d5868e75586210ec17b5ef5deddc276b460d16ae858c6624fd37aec316622a4060ca325Georgi Gerganovggerganov@gmail.com2025-06-09T12:57:58+03:00GitHubnoreply@github.com2025-06-09T12:57:58+03:00server : fix LRU check (#14079)
1485b460d16ae858c6624fd37aec316622a4060ca32591a8ee6a6f1f4c8547ff7b745ef95c6edc1d2af6Nicolò Scipionenicolo.scipione@codeplay.com2025-06-09T11:47:07+02:00GitHubnoreply@github.com2025-06-09T11:47:07+02:00sycl: Add reorder to Q6_K mmvq implementation (#13885)
148691a8ee6a6f1f4c8547ff7b745ef95c6edc1d2af6056eb74534ffd3efc50a9b854156eb6876a52a44Đinh Trọng Huy77562200+huydt84@users.noreply.github.com2025-06-09T13:15:31+09:00GitHubnoreply@github.com2025-06-09T05:15:31+01:00add geglu activation function (#14074)
1487056eb74534ffd3efc50a9b854156eb6876a52a44247e5c6e447707bb4539bdf1913d206088a8fc69Yuanhao Jijiyuanhao@apache.org2025-06-09T11:20:06+08:00GitHubnoreply@github.com2025-06-09T11:20:06+08:00CANN: Enable labeler for Ascend NPU (#13914)
1488247e5c6e447707bb4539bdf1913d206088a8fc695787b5da57e54dba760c2deeac1edf892e8fc450Diego Devesaslarengh@gmail.com2025-06-08T11:39:56-07:00GitHubnoreply@github.com2025-06-08T11:39:56-07:00cuda : fix buffer type check with integrated GPUs (#14069)
14895787b5da57e54dba760c2deeac1edf892e8fc450228f34c9ceefa3ea4f4d6933edd858121e8106cb吴小白296015668@qq.com2025-06-07T21:39:11+08:00GitHubnoreply@github.com2025-06-07T10:39:11-03:00ci: add LoongArch cross-compile build (#13944)
1490228f34c9ceefa3ea4f4d6933edd858121e8106cb0974ad7a7cd4bca846b15c484ff3be890135a52cAkarshan Biswasakarshan@menlo.ai2025-06-07T18:58:20+05:30GitHubnoreply@github.com2025-06-07T18:58:20+05:30SYCL: Implement few same quantized type copy kernels (#13739)
14910974ad7a7cd4bca846b15c484ff3be890135a52c745aa5319b9930068aff5e87cf5e9eef7227339bSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-07T14:13:12+02:00GitHubnoreply@github.com2025-06-07T14:13:12+02:00llama : fix llama_model_chat_template with template name (LLM_KV with suffix) (#14050)
1492745aa5319b9930068aff5e87cf5e9eef7227339b487a5e0401423bba02cd6e97e4d45131bb20b22bGeorgi Gerganovggerganov@gmail.com2025-06-06T14:11:15+03:00GitHubnoreply@github.com2025-06-06T14:11:15+03:00llama : deprecate llama_kv_self_ API (#14030)
1493487a5e0401423bba02cd6e97e4d45131bb20b22bd17a809ef0af09b16625e991a76f6fe80d9c332eGeorgi Gerganovggerganov@gmail.com2025-06-06T13:29:18+03:00GitHubnoreply@github.com2025-06-06T13:29:18+03:00context : fix SWA-related warning for multiple sequences (#14045)
149433fea9a8592fd338abe279b89350a3af2e0ff1f7806f05a353d322e97d97e02f910de2b408a8b4abYunzhe Jiayunzhe@calculet.tech2025-06-06T17:25:52+08:00Yunzhe Jiayunzhe@calculet.tech2025-06-06T17:35:48+08:00calrt: add load calbin
1495d17a809ef0af09b16625e991a76f6fe80d9c332e1caae7fc6c77551cb1066515e0f414713eebb367Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-06T09:03:25+02:00GitHubnoreply@github.com2025-06-06T09:03:25+02:00llama : support multiple classifier outputs and labels (#13940)
1496806f05a353d322e97d97e02f910de2b408a8b4abec9e0301fef6476df83e94842c3b625501c95566Yunzhe Jiayunzhe@calculet.tech2025-05-30T09:45:37+08:00Yunzhe Jiayunzhe@calculet.tech2025-06-06T09:24:26+08:00add calrt demo
14971caae7fc6c77551cb1066515e0f414713eebb367669c13e0f67edfba891c5bd7105eb4376bcf8626Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-05T17:42:31+02:00GitHubnoreply@github.com2025-06-05T17:42:31+02:00gguf-py : add add_classifier_output_labels method to writer (#14031)
1498669c13e0f67edfba891c5bd7105eb4376bcf8626146b88e8b3e16d22cea22f8be982a4d45e913b1eMasato Nakasakarillomas@gmail.com2025-06-05T23:00:29+09:00GitHubnoreply@github.com2025-06-05T16:00:29+02:00vulkan: Enable VK_KHR_cooperative_matrix extension for Intel Xe2 GPUs (#14001)
1499146b88e8b3e16d22cea22f8be982a4d45e913b1e7f37b6cf1e2c1b90bf0d9c8d91904b4b6c512748pockers21134406831+pockers21@users.noreply.github.com2025-06-05T06:25:29-07:00GitHubnoreply@github.com2025-06-05T16:25:29+03:00ci: fix CUDA build failure on autodl cloud machines (#14005)
15007f37b6cf1e2c1b90bf0d9c8d91904b4b6c5127483a077146a4761fdbd24bdd8eb098f46b8adc4ddaGeorgi Gerganovggerganov@gmail.com2025-06-05T15:29:22+03:00GitHubnoreply@github.com2025-06-05T15:29:22+03:00memory : migrate from llama_kv_cache to more generic llama_memory (#14006)
15013a077146a4761fdbd24bdd8eb098f46b8adc4ddad01d112abb055549d33bb8aac1755eb0c40918adDiego Devesaslarengh@gmail.com2025-06-05T02:57:42-07:00GitHubnoreply@github.com2025-06-05T11:57:42+02:00llama : allow using mmap without PrefetchVirtualMemory, apply GGML_WIN_VER to llama.cpp sources (#14013)
1502d01d112abb055549d33bb8aac1755eb0c40918ad9f47fa5792bae5312615439e68dbf1826913f7acOlexandr88radole1203@gmail.com2025-06-05T10:50:55+03:00GitHubnoreply@github.com2025-06-05T10:50:55+03:00readme : add badge (#13938)
15039f47fa5792bae5312615439e68dbf1826913f7ac9e31bec4fd53634c9e5b04650488a09a055f5dabSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-05T09:29:18+02:00GitHubnoreply@github.com2025-06-05T09:29:18+02:00vocab : warn about missing mask token (#14022)
15049e31bec4fd53634c9e5b04650488a09a055f5dab5a8ae3053ced350ed300ba91600519fcad1c6ba7Georgi Gerganovggerganov@gmail.com2025-06-05T09:06:29+03:00GitHubnoreply@github.com2025-06-05T09:06:29+03:00context : fix pos_min initialization upon error decode (#14008)
15055a8ae3053ced350ed300ba91600519fcad1c6ba70d3984424f2973c49c4bcabe4cc0153b4f90c601Jeff Bolzjbolz@nvidia.com2025-06-05T00:17:58-05:00GitHubnoreply@github.com2025-06-05T07:17:58+02:00vulkan: automatically deduce size of push constants (#13936)
15060d3984424f2973c49c4bcabe4cc0153b4f90c6013e63a58ef7addec35408e2eb67850d7cdc935dc3Ervin Áron Tasnádietasnadi@protonmail.com2025-06-04T22:02:00+02:00GitHubnoreply@github.com2025-06-04T22:02:00+02:00ggml-vulkan: adds support for op CONV_TRANSPOSE_1D (#13813)
15073e63a58ef7addec35408e2eb67850d7cdc935dc32589ad3704559f4dd860f5f303b19349c688a28aGeorgi Gerganovggerganov@gmail.com2025-06-04T18:58:20+03:00GitHubnoreply@github.com2025-06-04T18:58:20+03:00kv-cache : refactor the update/defrag mechanism (#13988)
15082589ad3704559f4dd860f5f303b19349c688a28a482548716f664f76e325ded58c9e8b7563e5e23aDiego Devesaslarengh@gmail.com2025-06-04T06:37:40-07:00GitHubnoreply@github.com2025-06-04T15:37:40+02:00ci : remove cuda 11.7 releases, switch runner to windows 2022 (#13997)
1509482548716f664f76e325ded58c9e8b7563e5e23a3ac67535c86e2fc43e4eddf594412acc370bbb04Diego Devesaslarengh@gmail.com2025-06-04T04:15:54-07:00GitHubnoreply@github.com2025-06-04T13:15:54+02:00releases : use dl backend for linux release, remove arm64 linux release (#13996)
15103ac67535c86e2fc43e4eddf594412acc370bbb040b4be4c435849b00dbd98b109cf7a22298d27b69Xuan-Son Nguyenson@huggingface.co2025-06-04T10:11:26+02:00GitHubnoreply@github.com2025-06-04T10:11:26+02:00llama-graph : use ggml_repeat_4d (#13998)
15110b4be4c435849b00dbd98b109cf7a22298d27b69e0e806f52ebcd0ee285c994fe8fd8b8787d2cb0aJohannes Gäßlerjohannesg@5d6.de2025-06-04T08:57:05+02:00GitHubnoreply@github.com2025-06-04T08:57:05+02:00CUDA: fix FTZ in FA for Gemma 3 (#13991)
1512e0e806f52ebcd0ee285c994fe8fd8b8787d2cb0a7e00e60ef86645a01fda738fef85b74afa016a34Georgi Gerganovggerganov@gmail.com2025-06-04T09:50:32+03:00GitHubnoreply@github.com2025-06-04T09:50:32+03:00kv-cache : fix unified::seq_rm to work with seq_id < 0 (#13985)
15137e00e60ef86645a01fda738fef85b74afa016a34ea1431b0fa3a8108aac1e0a94a13ccc4a749963eJeff Bolzjbolz@nvidia.com2025-06-03T13:30:22-05:00GitHubnoreply@github.com2025-06-03T20:30:22+02:00vulkan: fix warnings in perf logger querypool code (#13937)
151471e74a3ac929b8af91f16f73f3c2b9b2f796d207bfb1e012a0b7658e8f00ed4333d059943ea9d648lhezquic_lih@quicinc.com2025-06-02T16:54:58-07:00GitHubnoreply@github.com2025-06-02T16:54:58-07:00opencl: add `backend_synchronize` (#13939)
1515bfb1e012a0b7658e8f00ed4333d059943ea9d648363757628848a27a435bbf22ff9476e9aeda5f40rmatif66360289+rmatif@users.noreply.github.com2025-06-02T23:53:36ZGitHubnoreply@github.com2025-06-02T16:53:36-07:00OpenCL: Add concat, tsembd, upscale, tanh, pad and repeat (#13840)
1516363757628848a27a435bbf22ff9476e9aeda5f40ea394d7ab1f8101716d48ce9421c94c71b93a00fGeorgi Gerganovggerganov@gmail.com2025-06-02T21:34:40+03:00GitHubnoreply@github.com2025-06-02T21:34:40+03:00server : disable speculative decoding for SWA models (#13970)
1517ea394d7ab1f8101716d48ce9421c94c71b93a00f5582c49c3961269eca96822abfb87528e942dd07Georgi Gerganovggerganov@gmail.com2025-06-02T21:33:40+03:00GitHubnoreply@github.com2025-06-02T21:33:40+03:00metal : use F32 accumulators in FA kernels (#13975)
15185582c49c3961269eca96822abfb87528e942dd07c9bbc77931d223ed7e7cbcf1cb057bc02fd0db19Georgi Gerganovggerganov@gmail.com2025-06-02T20:54:26+03:00GitHubnoreply@github.com2025-06-02T20:54:26+03:00gemma : more consistent attention scaling for v2 and v3 (#13951)
1519c9bbc77931d223ed7e7cbcf1cb057bc02fd0db19bfd322796cd838f906535ff3352624fc46338894Olivier Chafikolivier.chafik@gmail.com2025-06-02T10:15:44-07:00GitHubnoreply@github.com2025-06-02T10:15:44-07:00`server`: update deepseek reasoning format (pass reasoning_content as diffs) (#13933)
1520bfd322796cd838f906535ff3352624fc46338894093e3f1feb16e25e58f7d61e01266c830dd424b8Xuan-Son Nguyenson@huggingface.co2025-06-02T16:29:28+02:00GitHubnoreply@github.com2025-06-02T16:29:28+02:00mtmd : fix memory leak in mtmd_helper_eval_chunk_single (#13961)
1521093e3f1feb16e25e58f7d61e01266c830dd424b8663445b0deb21fb602176da030d4154197a4fca6shalinib-ibmShalini.Salomi.Bodapati@ibm.com2025-06-02T17:48:36+05:30GitHubnoreply@github.com2025-06-02T15:18:36+03:00cmake : Handle mixed-case 'Power' strings in POWER CPU detection (#13966)
1522663445b0deb21fb602176da030d4154197a4fca67675c555a13c9f473249e59a54db35032ce8e0fcAtharva Dubeyatharva.dubey@codeplay.com2025-06-02T10:12:20+01:00GitHubnoreply@github.com2025-06-02T10:12:20+01:00sycl: quantize and reorder the input to q8_1 when reorder is enabled (#13826)
15237675c555a13c9f473249e59a54db35032ce8e0fc5e1c3aed4074480f63e914d6c44c93536ed1452aJohannes Gäßlerjohannesg@5d6.de2025-06-01T18:08:05+02:00GitHubnoreply@github.com2025-06-01T18:08:05+02:00gguf: fix failure on version == 0 (#13956)
15245e1c3aed4074480f63e914d6c44c93536ed1452ac496fe0b1da28618dd17bda8b0a6bf5004554080Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-01T18:07:21+02:00GitHubnoreply@github.com2025-06-01T18:07:21+02:00convert : fix nomic-bert-moe mask token (#13757)
1525c496fe0b1da28618dd17bda8b0a6bf5004554080e57bb87cede38341963a7a884630dbfb09c7dc00Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-06-01T17:23:11+02:00GitHubnoreply@github.com2025-06-01T17:23:11+02:00convert : fix vocab padding code for bert models (#13954)
1526e57bb87cede38341963a7a884630dbfb09c7dc00f3a4b1659ccc94ebdf3590d472b518377bfecc7aAaron Teoaaron.teo1@ibm.com2025-06-01T22:53:57+08:00GitHubnoreply@github.com2025-06-01T16:53:57+02:00ggml: check if non-native endian model is being loaded (#13943)
1527f3a4b1659ccc94ebdf3590d472b518377bfecc7a108009f5c7267b77292c11f788f602d6d7ae8fcdGeorgi Gerganovggerganov@gmail.com2025-06-01T12:23:14+03:00Georgi Gerganovggerganov@gmail.com2025-06-01T13:43:57+03:00sync : ggml
1528108009f5c7267b77292c11f788f602d6d7ae8fcdd337252acf14a91a685c355fa4f3f599a8068207Kai Pastordg0yt@darc.de2025-05-31T12:49:55+02:00Georgi Gerganovggerganov@gmail.com2025-06-01T13:43:57+03:00vulkan : Remove unexpected ; (ggml/1253)
1529d337252acf14a91a685c355fa4f3f599a8068207af6f91db470da543dc32bc00c057aad8f060dfdbKai Pastordg0yt@darc.de2025-05-31T12:39:19+02:00Georgi Gerganovggerganov@gmail.com2025-06-01T13:43:57+03:00cmake : Fix broken CMake error messages (ggml/1252)
1530af6f91db470da543dc32bc00c057aad8f060dfdba7b8d35f780ab3e7639ae5345442fb1ad10f0163Radoslav Gerganovrgerganov@gmail.com2025-05-30T09:11:09+03:00Georgi Gerganovggerganov@gmail.com2025-06-01T13:43:57+03:00ggml : remove ggml_graph_import and ggml_graph_export declarations (ggml/1247)
1531a7b8d35f780ab3e7639ae5345442fb1ad10f01636eba72b71c677ce9aae33c422a6e67008137987aGeorgi Gerganovggerganov@gmail.com2025-05-29T13:29:50+03:00Georgi Gerganovggerganov@gmail.com2025-06-01T13:43:57+03:00sync : whisper.cpp (ggml/1250)
15326eba72b71c677ce9aae33c422a6e67008137987afedf034a98378059274e4243d2a370a243335e73Radoslav Gerganovrgerganov@gmail.com2025-05-29T08:34:46+03:00Georgi Gerganovggerganov@gmail.com2025-06-01T13:43:57+03:00ggml : install dynamic backends (ggml/1240)
1533fedf034a98378059274e4243d2a370a243335e738726392d3d927fe3e504868d3548d694496ee37eDaniel Tangdanielzgtg.opensource@gmail.com2025-05-27T20:58:46-04:00Georgi Gerganovggerganov@gmail.com2025-06-01T13:43:57+03:00ggml : Print backtrace on uncaught C++ exceptions (ggml/1232)
15348726392d3d927fe3e504868d3548d694496ee37ec04621711a893cbd09cff6c927cb005bc6749e36ddh0chemist-mulches-39@icloud.com2025-06-01T03:44:30-05:00GitHubnoreply@github.com2025-06-01T11:44:30+03:00readme : update bindings (#13950)
1535c04621711a893cbd09cff6c927cb005bc6749e360fc16b42e8793364918e830c234c1f3caec29daeGeorgi Gerganovggerganov@gmail.com2025-06-01T11:42:16+03:00GitHubnoreply@github.com2025-06-01T11:42:16+03:00parallel : fix n_junk == 0 (#13952)
15360fc16b42e8793364918e830c234c1f3caec29dae053b1539c02617eff744f89525ee57497c3c1fbeGeorgi Gerganovggerganov@gmail.com2025-06-01T11:39:27+03:00GitHubnoreply@github.com2025-06-01T11:39:27+03:00kv-cache : split implementation in separate sources (#13920)
1537053b1539c02617eff744f89525ee57497c3c1fbeb3a89c3d9e34c28c5be70d8b687a84775746d4a0Max Krasnyanskyquic_maxk@quicinc.com2025-05-31T15:39:19-07:00GitHubnoreply@github.com2025-05-31T15:39:19-07:00threading: support for GGML_SCHED_PRIO_LOW, update thread info on Windows to avoid throttling (#12995)
1538b3a89c3d9e34c28c5be70d8b687a84775746d4a0e15898d1c7e87f1b3d14e0a3c385eeb424b085feJiří Podivín66251151+jpodivin@users.noreply.github.com2025-05-31T18:58:35+02:00GitHubnoreply@github.com2025-05-31T18:58:35+02:00docs : Note about necessity of having libcurl installed for standard build. (#13945)
1539e15898d1c7e87f1b3d14e0a3c385eeb424b085fe803f8baf4f741d2f0465c46c33f285886b97a071Olivier Chafikolivier.chafik@gmail.com2025-05-31T08:26:10-07:00GitHubnoreply@github.com2025-05-31T08:26:10-07:00server: allow unclosed thinking tags (#13931)
1540803f8baf4f741d2f0465c46c33f285886b97a0713600cc2886956fc0a07ef6ad2f4128ccfdbc8c6fGeorgi Gerganovggerganov@gmail.com2025-05-31T15:58:33+03:00GitHubnoreply@github.com2025-05-31T15:58:33+03:00llama : deprecate explicit kv_self defrag/update calls (#13921)
15413600cc2886956fc0a07ef6ad2f4128ccfdbc8c6fc7e0a2054b908c28bf93bb18d4b63ccbff2c4127Georgi Gerganovggerganov@gmail.com2025-05-31T15:57:44+03:00GitHubnoreply@github.com2025-05-31T15:57:44+03:00llama : use n_swa + n_ubatch cells for SWA cache (#13833)
1542c7e0a2054b908c28bf93bb18d4b63ccbff2c41273f55f781f1da9241f209648636fa0426fe62a495igardev49397134+igardev@users.noreply.github.com2025-05-31T12:56:08+03:00GitHubnoreply@github.com2025-05-31T11:56:08+02:00webui : Replace alert and confirm with custom modals. (#13711)
15433f55f781f1da9241f209648636fa0426fe62a49551fa76f172957c9916e43aeb581f82c533e12394Georgi Gerganovggerganov@gmail.com2025-05-31T12:55:57+03:00GitHubnoreply@github.com2025-05-31T12:55:57+03:00llama : auto-batch preparation (#13845)
154451fa76f172957c9916e43aeb581f82c533e1239412d0188c0dc6146ffde6d277a93f232ccbe699f8Xuan-Son Nguyenson@huggingface.co2025-05-31T10:14:29+02:00GitHubnoreply@github.com2025-05-31T10:14:29+02:00mtmd : drop `_shared` from `libmtmd` name, merge helpers into libmtmd (⚠️ breaking change) (#13917)
154512d0188c0dc6146ffde6d277a93f232ccbe699f8eb3949938e82a128855bc0676220bb2ce6e4228dGeorgi Gerganovggerganov@gmail.com2025-05-31T10:24:04+03:00GitHubnoreply@github.com2025-05-31T10:24:04+03:00kv-cache : refactor + add llama_memory_state_i (#13746)
1546eb3949938e82a128855bc0676220bb2ce6e4228de562eece7cb476276bfc4cbb18deb7c0369b2233Shawn yang137684499+Yangxiaoz@users.noreply.github.com2025-05-31T14:48:04+08:00GitHubnoreply@github.com2025-05-31T08:48:04+02:00CUDA: add a prop in ggml_cuda_device_infor for distinguish iGPU or dGPU in cuda (#13856) (#13895)
1547e562eece7cb476276bfc4cbb18deb7c0369b2233b47ab7b8e9c4f6154eb6abb6234866ab117d64c7Johannes Gäßlerjohannesg@5d6.de2025-05-30T21:22:03+02:00GitHubnoreply@github.com2025-05-30T21:22:03+02:00CUDA: fix typo in FlashAttention code (#13926)
1548b47ab7b8e9c4f6154eb6abb6234866ab117d64c7dd665cc9d4b378626cde11ea0c4c91f514fdc994Diego Devesaslarengh@gmail.com2025-05-30T09:56:19-07:00GitHubnoreply@github.com2025-05-30T18:56:19+02:00sched : avoid changing cur_copy when a graph is already allocated (#13922)
1549dd665cc9d4b378626cde11ea0c4c91f514fdc994df0c0c7d02f951dc639d48fd536f79200460ac83Georgi Gerganovggerganov@gmail.com2025-05-30T19:38:07+03:00GitHubnoreply@github.com2025-05-30T19:38:07+03:00parallel : increase the variability of the prompt lengths (#13927)
1550df0c0c7d02f951dc639d48fd536f79200460ac83b49a8ff96b769b8a4c36d89fb783ec0135be582bDiego Devesaslarengh@gmail.com2025-05-30T07:37:18-07:00GitHubnoreply@github.com2025-05-30T16:37:18+02:00cuda : prevent using split buffers with 3d/4d matrices (#13919)
1551b49a8ff96b769b8a4c36d89fb783ec0135be582b53f925074de02c5304b00c14b4d6d8910c58667dAkarshan Biswasakarshan@menlo.ai2025-05-30T19:40:57+05:30GitHubnoreply@github.com2025-05-30T19:40:57+05:30SYCL: Add mrope kernel (#13755)
155253f925074de02c5304b00c14b4d6d8910c58667ddb38704f0133be7832123495fa8fc2601ea999d4Georgi Gerganovggerganov@gmail.com2025-05-30T16:25:45+03:00GitHubnoreply@github.com2025-05-30T16:25:45+03:00sync : vendor (#13901)
1553db38704f0133be7832123495fa8fc2601ea999d407e4351ce663a7802b31f92fd7bc0e555c2044b6Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-30T14:50:43+02:00GitHubnoreply@github.com2025-05-30T14:50:43+02:00convert : fix rwkv bos/eos token (#13844)
155407e4351ce663a7802b31f92fd7bc0e555c2044b6291f2b6913c7ef8350dbf0e77da38f7af131a08eXuan-Son Nguyenson@huggingface.co2025-05-30T12:24:37+02:00GitHubnoreply@github.com2025-05-30T12:24:37+02:00convert : allow partial update to the chkhsh pre-tokenizer list (#13847)
1555291f2b6913c7ef8350dbf0e77da38f7af131a08e2c90da4c7ec694797f524042aaafbb047a7e65ffĐinh Trọng Huy77562200+huydt84@users.noreply.github.com2025-05-30T18:56:02+09:00GitHubnoreply@github.com2025-05-30T11:56:02+02:00llama : add support for DistilBert (#13907)
15562c90da4c7ec694797f524042aaafbb047a7e65ffec9e0301fef6476df83e94842c3b625501c95566zhangkaihuozhangkaihuo@gmail.com2025-05-30T16:31:48+08:00GitHubnoreply@github.com2025-05-30T10:31:48+02:00llama : use llm_build_granite for minicpm (#13911)
1557ec9e0301fef6476df83e94842c3b625501c95566e83ba3e460651b20a594e9f2f0f0bffb998d3ce1Christian Kastnerckk@kvr.at2025-05-30T01:28:54+02:00GitHubnoreply@github.com2025-05-30T01:28:54+02:00cmake: Guard GGML_CPU_ALL_VARIANTS by architecture (#13890)
1558e83ba3e460651b20a594e9f2f0f0bffb998d3ce12b131621e60d8ec2cc961201beb6773ab37b6b69Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-29T21:42:31+02:00GitHubnoreply@github.com2025-05-29T21:42:31+02:00llama : add support for jina-reranker-v2 (#13900)
15592b131621e60d8ec2cc961201beb6773ab37b6b6954a2c7a8cd8a32b44e3a98c2999b0f5c9114be5cSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-29T15:36:05+02:00GitHubnoreply@github.com2025-05-29T15:36:05+02:00gguf-py : add support for sub_type (in arrays) in GGUFWriter add_key_value method (#13561)
156054a2c7a8cd8a32b44e3a98c2999b0f5c9114be5c21fcc21ad5e5de2daa5da8d08dbbcc86b8d815d7Yibo Caiyibo.cai@arm.com2025-05-29T19:39:20+08:00GitHubnoreply@github.com2025-05-29T14:39:20+03:00arm64: optimize q4_k_q8_k kernel with i8mm (#13886)
156121fcc21ad5e5de2daa5da8d08dbbcc86b8d815d7dd8ba93416f492c168f7e20f0b1434c08ebeaeb5Christian Kastnerckk@kvr.at2025-05-29T12:50:25+02:00GitHubnoreply@github.com2025-05-29T12:50:25+02:00cmake: Factor out CPU architecture detection (#13883)
1562dd8ba93416f492c168f7e20f0b1434c08ebeaeb566c92061f5c34d2f9a630b7b50cca5ce3ebb4b16Vineel Abhinav131174187+vineelabhinav@users.noreply.github.com2025-05-29T14:48:43+05:30GitHubnoreply@github.com2025-05-29T12:18:43+03:00ggml: aarch64: Implement SVE F32 kernels for Mamba Sequential Scan Algorithm (#13882)
156366c92061f5c34d2f9a630b7b50cca5ce3ebb4b165ca82fc1d7f8cb27f3cbb3019e944a80d5219828Georgi Gerganovggerganov@gmail.com2025-05-29T12:17:16+03:00GitHubnoreply@github.com2025-05-29T12:17:16+03:00tests : remove json.hpp from a test (#13880)
15645ca82fc1d7f8cb27f3cbb3019e944a80d52198286385b843a8dc8e15b8362196039720c58dd79fa2Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-29T10:00:57+02:00GitHubnoreply@github.com2025-05-29T10:00:57+02:00convert : workaround for AutoConfig dummy labels (#13881)
15656385b843a8dc8e15b8362196039720c58dd79fa21b8fb8152d0f3357a9c1d9b4c02f6cc5b9cb7232Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-29T08:15:01+02:00GitHubnoreply@github.com2025-05-29T08:15:01+02:00llama : add RobertaForSequenceClassification reranker support (#13875)
15661b8fb8152d0f3357a9c1d9b4c02f6cc5b9cb723253ae30640e131082d8d19bd80485b47c4553d551Vineel Abhinav131174187+vineelabhinav@users.noreply.github.com2025-05-29T11:31:33+05:30GitHubnoreply@github.com2025-05-29T09:01:33+03:00ggml: aarch64: Implement SVE F32 kernels for vector functions (#13843)
156753ae30640e131082d8d19bd80485b47c4553d551763d06edb7dd5094ea58bad1d81e2e8d35033e34Beinsezii39478211+Beinsezii@users.noreply.github.com2025-05-28T14:50:20-07:00GitHubnoreply@github.com2025-05-28T23:50:20+02:00gguf-py : fix SafetensorRemote return on undefined size (< 0) (#13841)
1568763d06edb7dd5094ea58bad1d81e2e8d35033e3410961339b26bd2eff01d5479e8879f435da261b7Xuan-Son Nguyenson@huggingface.co2025-05-28T22:35:31+02:00GitHubnoreply@github.com2025-05-28T22:35:31+02:00llama : fix KV shift for qwen2vl (#13870)
156910961339b26bd2eff01d5479e8879f435da261b7d98f2a35fcf4a8d3e660ad48cd19e2a1f3d5b2efXuan-Son Nguyenson@huggingface.co2025-05-28T22:35:22+02:00GitHubnoreply@github.com2025-05-28T22:35:22+02:00mtmd : move helpers to dedicated library (⚠️ breaking change) (#13866)
1570d98f2a35fcf4a8d3e660ad48cd19e2a1f3d5b2efe0e3aa231d899720c2864d09cdb89d4c400eeb55bandoti141645996+bandoti@users.noreply.github.com2025-05-28T15:46:47-03:00GitHubnoreply@github.com2025-05-28T15:46:47-03:00ci: disable LLAMA_CURL for Linux cross-builds (#13871)
1571e0e3aa231d899720c2864d09cdb89d4c400eeb55aa6dff05be25709bb218bf648951d690029c4b19Đinh Trọng Huy77562200+huydt84@users.noreply.github.com2025-05-29T02:01:58+09:00GitHubnoreply@github.com2025-05-28T19:01:58+02:00llama : add support for BertForSequenceClassification reranker (#13858)
1572aa6dff05be25709bb218bf648951d690029c4b19c962ae3382a1e759c8517a229549ee53685313a1Đinh Trọng Huy77562200+huydt84@users.noreply.github.com2025-05-28T23:34:18+09:00GitHubnoreply@github.com2025-05-28T16:34:18+02:00convert: small addition to support LlamaModel (#13838)
1573c962ae3382a1e759c8517a229549ee53685313a1a3938fb53d0b5183db4f5a6db21fd9122a0e4780SkyIflyinskyin2013@gmail.com2025-05-28T22:33:54+08:00GitHubnoreply@github.com2025-05-28T16:33:54+02:00server: fix remove 'image_url'/'input_audio' json-object effectlly for 'llama_params' in multimodal-model-mode (#13853)
1574a3938fb53d0b5183db4f5a6db21fd9122a0e4780f7873fc698c09047e2873630ab7e7730a0bfb224Xuan-Son Nguyenson@huggingface.co2025-05-28T16:12:35+02:00GitHubnoreply@github.com2025-05-28T16:12:35+02:00convert : fix qwen omni conversion (#13859)
1575f7873fc698c09047e2873630ab7e7730a0bfb224a68247439bd6fb756cc93ad2817e55a02aa0b100Alex Fanthomexfanth@gmail.com2025-05-28T14:49:28+01:00GitHubnoreply@github.com2025-05-28T15:49:28+02:00tests : change umlaut test (#11600)
1576a68247439bd6fb756cc93ad2817e55a02aa0b10026b79b6cb3e7840ff15729350e95907e19f9f480Johannes Gäßlerjohannesg@5d6.de2025-05-28T13:33:37+02:00GitHubnoreply@github.com2025-05-28T13:33:37+02:00CUDA: fix FA tg at long context for CC >= 8.9 (#13852)
157726b79b6cb3e7840ff15729350e95907e19f9f4801e8659e65ad0f640674d2785d02b556ab938728cXuan-Son Nguyenson@huggingface.co2025-05-28T10:05:54+02:00GitHubnoreply@github.com2025-05-28T10:05:54+02:00convert : fix tensor naming conflict for llama 4 vision (#13836)
15781e8659e65ad0f640674d2785d02b556ab938728ca3c30846e410c91c11d7bf80978795a03bb03deeleo-ponynengjunma@outlook.com2025-05-28T11:54:20+08:00GitHubnoreply@github.com2025-05-28T11:54:20+08:00CANN: Add SOC TYPE printing in cmake configuration (#13837)
1579a3c30846e410c91c11d7bf80978795a03bb03dee1701d4c54f93c0d203bf92ea7202a94b037c2338lhezquic_lih@quicinc.com2025-05-27T12:56:08-07:00GitHubnoreply@github.com2025-05-27T12:56:08-07:00opencl: add new ops - `argsort`, `div`, `sub`, `addrows`, `sigmoid`, `group_norm` (#13787)
15801701d4c54f93c0d203bf92ea7202a94b037c2338bef817638780dcbd8c0c80c97b7a4f8e92c8fe74lhezquic_lih@quicinc.com2025-05-27T12:53:14-07:00GitHubnoreply@github.com2025-05-27T12:53:14-07:00opencl: mark `mul_mat` `f32f32` as supporting non-contiguous tensors (#13790)
1581bef817638780dcbd8c0c80c97b7a4f8e92c8fe7434b7c0439ed0f98575cc4689dfecd98991dee8beJeff Bolzjbolz@nvidia.com2025-05-27T11:39:07-05:00GitHubnoreply@github.com2025-05-27T18:39:07+02:00vulkan: use timestamp queries for GGML_VULKAN_PERF (#13817)
158234b7c0439ed0f98575cc4689dfecd98991dee8bef3101a8cc665f73217c752a10a7042889275cfbcGeorgi Gerganovggerganov@gmail.com2025-05-27T19:08:44+03:00GitHubnoreply@github.com2025-05-27T19:08:44+03:00cmake : add llama-cparams.cpp to build (#13832)
1583f3101a8cc665f73217c752a10a7042889275cfbc1c49c70d07ef87635daa5e8fdd0b5bfd88493dd3Akarshan Biswasakarshan@menlo.ai2025-05-27T20:52:59+05:30GitHubnoreply@github.com2025-05-27T20:52:59+05:30SYCL: add gelu_erf kernel (#13749)
15841c49c70d07ef87635daa5e8fdd0b5bfd88493dd3a8ea03d8ad9c984fe6cfafead183ab188f8cbeb0Georgi Gerganovggerganov@gmail.com2025-05-27T18:04:38+03:00Georgi Gerganovggerganov@gmail.com2025-05-27T18:05:33+03:00sync : ggml
1585a8ea03d8ad9c984fe6cfafead183ab188f8cbeb005f6ac6283a7859a03cd59405504262c85c4bf06Xuan-Son Nguyenson@huggingface.co2025-05-27T15:53:55+02:00GitHubnoreply@github.com2025-05-27T15:53:55+02:00ggml : add ggml_repeat_4d (#13824)
158605f6ac6283a7859a03cd59405504262c85c4bf06bc583e3c63c04a11d287c108ea9e6a515ead0423xctanxc-tan@outlook.com2025-05-27T21:21:36+08:00GitHubnoreply@github.com2025-05-27T16:21:36+03:00ggml : riscv: add xtheadvector support (#13720)
1587bc583e3c63c04a11d287c108ea9e6a515ead042372b090da2c50e540143fd312a2f9aa5f151e6136Xuan-Son Nguyenson@huggingface.co2025-05-27T14:06:10+02:00GitHubnoreply@github.com2025-05-27T14:06:10+02:00mtmd : support Qwen 2.5 Omni (input audio+vision, no audio output) (#13784)
158872b090da2c50e540143fd312a2f9aa5f151e61367fe03e7446ed4388539d3bcc013ae4e851b8d1e2bandoti141645996+bandoti@users.noreply.github.com2025-05-27T08:52:40-03:00GitHubnoreply@github.com2025-05-27T08:52:40-03:00docs: remove link for llama-cli function calling (#13810)
15897fe03e7446ed4388539d3bcc013ae4e851b8d1e2952f3953c1b61cc70e79e536c42ddce6a5ea5ea7Christian Kastnerckk@kvr.at2025-05-27T13:18:39+02:00GitHubnoreply@github.com2025-05-27T13:18:39+02:00ggml-cpu: x86 feature detection is specific to x86 (#13811)
1590952f3953c1b61cc70e79e536c42ddce6a5ea5ea781713121ee56755ba4a147d1a1e3fa248ec31a66Diego Devesaslarengh@gmail.com2025-05-27T04:05:18-07:00GitHubnoreply@github.com2025-05-27T13:05:18+02:00ggml : allow CUDA graphs when using pipeline parallelism (#13814)
159181713121ee56755ba4a147d1a1e3fa248ec31a66f9cd68398baf2ba8af4725ca9ed00bef205e6706Georgi Gerganovggerganov@gmail.com2025-05-27T13:49:41+03:00GitHubnoreply@github.com2025-05-27T13:49:41+03:00kv-cells : track min/max used cells and per-sequence positions (#13808)
1592f9cd68398baf2ba8af4725ca9ed00bef205e67064f81b33e324b1669b282eb81104e4a1131be7dceGeorgi Gerganovggerganov@gmail.com2025-05-27T12:07:52+03:00GitHubnoreply@github.com2025-05-27T12:07:52+03:00sampling : make sure samplers return at least 1 token (#13822)
15934f81b33e324b1669b282eb81104e4a1131be7dcecdf94a18023c92f41808ec874ba577d914674717Georgi Gerganovggerganov@gmail.com2025-05-27T09:40:59+03:00GitHubnoreply@github.com2025-05-27T09:40:59+03:00llama : validate seq id batch input (#13809)
1594cdf94a18023c92f41808ec874ba577d914674717a26c4cc11ec7c6574e3691e90ecdbd67deeea35bOlivier Chafikolivier.chafik@gmail.com2025-05-26T14:34:27-07:00GitHubnoreply@github.com2025-05-26T22:34:27+01:00server: --offline mode (#13804)
1595a26c4cc11ec7c6574e3691e90ecdbd67deeea35b4265a87b59ebfc25f35adbf4db3b608995b0a78aGeorgi Gerganovggerganov@gmail.com2025-05-26T22:24:01+03:00GitHubnoreply@github.com2025-05-26T22:24:01+03:00scripts : add option to compare commits in Debug (#13806)
15964265a87b59ebfc25f35adbf4db3b608995b0a78a6f180b915c9ed9ec0c240b5dcd64644988fb5e82Georgi Gerganovggerganov@gmail.com2025-05-26T22:14:52+03:00GitHubnoreply@github.com2025-05-26T22:14:52+03:00cuda : avoid cuGetErrorString (#13791)
15976f180b915c9ed9ec0c240b5dcd64644988fb5e8203f582ae8fccecff225c30a2802461b44761e822Akarshan Biswasakarshan@menlo.ai2025-05-26T21:10:36+05:30GitHubnoreply@github.com2025-05-26T21:10:36+05:30SYCL: Add non contiguous support in RMS_NORM and NORM kernels (#13611)
159803f582ae8fccecff225c30a2802461b44761e82288c125f2acce0e25e5fc8481ab0681415fc64a10Olivier Chafikolivier.chafik@gmail.com2025-05-26T08:03:57-07:00GitHubnoreply@github.com2025-05-26T16:03:57+01:00server: fix streaming crashes (#13786)
159988c125f2acce0e25e5fc8481ab0681415fc64a10d74e94c1b3ac02db01e47008e1f8d371029d81acstandby24x7standby24x7@gmail.com2025-05-26T23:55:24+09:00GitHubnoreply@github.com2025-05-26T16:55:24+02:00examples/training: Fix file name in README (#13803)
1600d74e94c1b3ac02db01e47008e1f8d371029d81acf13847cfb560d92492b480cca2c5d6aa9473cde3Olivier Chafikolivier.chafik@gmail.com2025-05-26T06:56:49-07:00GitHubnoreply@github.com2025-05-26T14:56:49+01:00`server`: fix format of streamed tool call deltas (diff name, fix id location) (#13800)
1601f13847cfb560d92492b480cca2c5d6aa9473cde379c137f77677b3c8ee3c60a7da033721b938399aOlivier Chafikolivier.chafik@gmail.com2025-05-26T06:16:37-07:00GitHubnoreply@github.com2025-05-26T14:16:37+01:00server: fix regression on streamed non-chat completion w/ stops (#13785)
160279c137f77677b3c8ee3c60a7da033721b938399a22229314fc46b2f741bb21b12cde71f6c6a60b52Georgi Gerganovggerganov@gmail.com2025-05-26T14:03:54+03:00GitHubnoreply@github.com2025-05-26T14:03:54+03:00examples : allow extracting embeddings from decoder contexts (#13797)
160322229314fc46b2f741bb21b12cde71f6c6a60b529012eb9b454a82eaa4cd77ae904c0ea391e4db42Georgi Gerganovggerganov@gmail.com2025-05-26T12:57:50+03:00GitHubnoreply@github.com2025-05-26T12:57:50+03:00llama : clarify deprecation message (#13794)
16049012eb9b454a82eaa4cd77ae904c0ea391e4db42fef693dc6b959a8e8ba11558fbeaad0b264dd457Romain Biessyromain.biessy@codeplay.com2025-05-26T10:28:53+02:00GitHubnoreply@github.com2025-05-26T10:28:53+02:00sycl: Add more debug prints (#13640)
1605fef693dc6b959a8e8ba11558fbeaad0b264dd4572d38b6e4004fb1c341723e657fb1e71a4d3fb473Jeff Bolzjbolz@nvidia.com2025-05-25T23:02:07-05:00GitHubnoreply@github.com2025-05-26T06:02:07+02:00vulkan: mark IM2COL as supporting non-contig (#13783)
16062d38b6e4004fb1c341723e657fb1e71a4d3fb473e121edc4324a640be11b7e567edd39b721b0f8e4Bizhao Shi37729561+shibizhao@users.noreply.github.com2025-05-26T10:20:18+08:00GitHubnoreply@github.com2025-05-26T10:20:18+08:00CANN: Add the basic supports of Flash Attention kernel (#13627)
1607e121edc4324a640be11b7e567edd39b721b0f8e42f099b510f460374acd52742b494595e3e3442d3Olivier Chafikolivier.chafik@gmail.com2025-05-26T00:30:51+01:00GitHubnoreply@github.com2025-05-26T00:30:51+01:00`server`: add `--reasoning-budget 0` to disable thinking (incl. qwen3 w/ enable_thinking:false) (#13771)
16082f099b510f460374acd52742b494595e3e3442d3aa50ba462f63759e1731227f20f5009cfc2a0f16Xuan-Son Nguyenson@huggingface.co2025-05-25T19:02:18+02:00GitHubnoreply@github.com2025-05-25T18:02:18+01:00webui : bump max upload file size to 500MB (#13779)
1609aa50ba462f63759e1731227f20f5009cfc2a0f16de2ef53a4b2c0d703749a309d19fe68fd8f1b9acSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-25T16:22:29+02:00GitHubnoreply@github.com2025-05-25T16:22:29+02:00tests : improve UGM tokenizer test coverage (#13773)
1610de2ef53a4b2c0d703749a309d19fe68fd8f1b9acc508256db2de2b032e19c8ed833f4683c827c9a1Georgi Gerganovggerganov@gmail.com2025-05-25T16:34:36+03:00GitHubnoreply@github.com2025-05-25T16:34:36+03:00kv-cache : rework kv_cell (#13706)
1611c508256db2de2b032e19c8ed833f4683c827c9a140aaa8a403df5dcfb515eb2fd7a817fd99779526Percy Piperpiper.percy@googlemail.com2025-05-25T13:35:53+01:00GitHubnoreply@github.com2025-05-25T15:35:53+03:00rpc : Fix build on OpenBSD (#13541)
161240aaa8a403df5dcfb515eb2fd7a817fd99779526a08c1d2845dc279d58d826ef3c3ecad97cbbcef7Xuan-Son Nguyenson@huggingface.co2025-05-25T14:06:32+02:00GitHubnoreply@github.com2025-05-25T14:06:32+02:00mtmd : add support for Qwen2-Audio and SeaLLM-Audio (#13760)
1613a08c1d2845dc279d58d826ef3c3ecad97cbbcef7d785f9c1fd1a1929c6d0e2a0b12cae5db867908bddpasa112642920+ddpasa@users.noreply.github.com2025-05-25T14:04:49+02:00GitHubnoreply@github.com2025-05-25T14:04:49+02:00docs : add Moondream2 pre-quantized link (#13745)
1614d785f9c1fd1a1929c6d0e2a0b12cae5db867908b4032ca406632212b075e3c61c2a4476128321410Olivier Chafikolivier.chafik@gmail.com2025-05-25T10:45:49+01:00GitHubnoreply@github.com2025-05-25T10:45:49+01:00server: fix/test add_generation_prompt (#13770)
16154032ca406632212b075e3c61c2a4476128321410515fdbf7ed839dfe6a24aeb6225936609a7f6d6dPiotr Jasiukajtisestibi@me.com2025-05-25T10:29:43+02:00GitHubnoreply@github.com2025-05-25T10:29:43+02:00llama : add support for Qwen3 MoE tied word embeddings (#13768)
1616f5cd27b71da3ac375a04a41643d14fc779a8057ba2d02d5793fd9af7a7224773456501691b95fd02Olivier Chafikolivier.chafik@gmail.com2025-05-25T01:48:08+01:00GitHubnoreply@github.com2025-05-25T01:48:08+01:00`server`: streaming of tool calls and thoughts when `--jinja` is on (#12379)
1617a2d02d5793fd9af7a7224773456501691b95fd0217fc817b58942569c43aa63299cedde217b61f68Diego Devesaslarengh@gmail.com2025-05-24T15:55:16-07:00GitHubnoreply@github.com2025-05-25T00:55:16+02:00releases : bundle llvm omp library in windows release (#13763)
161817fc817b58942569c43aa63299cedde217b61f682bd1b30f6979235ec67b95c183b9b77baa7ab9ceDiego Devesaslarengh@gmail.com2025-05-24T13:27:03-07:00GitHubnoreply@github.com2025-05-24T22:27:03+02:00releases : enable openmp in windows cpu backend build (#13756)
16192bd1b30f6979235ec67b95c183b9b77baa7ab9ce259469c4b57c1a32606353bcac52ba683424a990Diego Devesaslarengh@gmail.com2025-05-24T13:26:47-07:00GitHubnoreply@github.com2025-05-24T22:26:47+02:00ggml-cpu : set openmp wait time if not set (#13758)
1620259469c4b57c1a32606353bcac52ba683424a9904c32832c59e97d96c19349fdc92763e8949fda830cc4mpicard12@live.de2025-05-24T16:49:12+02:00GitHubnoreply@github.com2025-05-24T16:49:12+02:00Move GLM4 f32 attention fix to the correct function (#13750)
16214c32832c59e97d96c19349fdc92763e8949fda83c3a2624339187e89c4f65fd72a5fe7103968b5adXuan-Son Nguyenson@huggingface.co2025-05-24T13:06:47+02:00GitHubnoreply@github.com2025-05-24T13:06:47+02:00ggml : add ggml_gelu_erf() CUDA kernel (#13719)
1622c3a2624339187e89c4f65fd72a5fe7103968b5adffd0eae60b7642e942c8245b89642527e36099e6Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-24T12:29:09+02:00GitHubnoreply@github.com2025-05-24T12:29:09+02:00vocab : fix ugm tokenizer precision (#13743)
1623ffd0eae60b7642e942c8245b89642527e36099e6b775345d788ac16260e7eef49e11fe57ee5677f7Johannes Gäßlerjohannesg@5d6.de2025-05-24T11:46:19+02:00GitHubnoreply@github.com2025-05-24T11:46:19+02:00CUDA: fix race condition in FA vector kernels (#13742)
1624b775345d788ac16260e7eef49e11fe57ee5677f7a70a8a69c2ad3de8d5525ad2b185b098011be2e8Diego Devesaslarengh@gmail.com2025-05-23T13:14:00-07:00GitHubnoreply@github.com2025-05-23T23:14:00+03:00ci : enable winget package updates (#13734)
1625a70a8a69c2ad3de8d5525ad2b185b098011be2e8d13d0f6135803822ec1cd7e3efb49360b88a1bdfDiego Devesaslarengh@gmail.com2025-05-23T13:09:38-07:00GitHubnoreply@github.com2025-05-23T22:09:38+02:00ci : add winget package updater (#13732)
1626d13d0f6135803822ec1cd7e3efb49360b88a1bdf8a2afb7520bbc8f9fa1bbe314d5f2807eb0116b2Georgi Gerganovggerganov@gmail.com2025-05-23T20:16:13+03:00GitHubnoreply@github.com2025-05-23T20:16:13+03:00hparams : initialize arrays (#13728)
16278a2afb7520bbc8f9fa1bbe314d5f2807eb0116b29ecf3e66a38f20e41d221f62f05b17f70d040839Xuan-Son Nguyenson@huggingface.co2025-05-23T17:07:04+02:00GitHubnoreply@github.com2025-05-23T17:07:04+02:00llama : allow custom list of swa_layers (#13726)
16289ecf3e66a38f20e41d221f62f05b17f70d040839faaaff5f947064d13ef8b98659d81a1384c3e57bXuan-Son Nguyenson@huggingface.co2025-05-23T11:03:47+02:00GitHubnoreply@github.com2025-05-23T11:03:47+02:00server : support audio input (#13714)
1629faaaff5f947064d13ef8b98659d81a1384c3e57be16c4731c7a6bfa8f61b5b39c77245a37e7fc232Chenguang Li757486878@qq.com2025-05-23T16:47:53+08:00GitHubnoreply@github.com2025-05-23T16:47:53+08:00CANN: Support MUL_MAT_ID for q8_0 and q4_0 (#13705)
1630e16c4731c7a6bfa8f61b5b39c77245a37e7fc2321dcd01960c33f52afb782b3d850fc2149a08cc6bXuan-Son Nguyenson@huggingface.co2025-05-23T08:12:48+02:00GitHubnoreply@github.com2025-05-23T08:12:48+02:00ggml : fix the order of ggml_unary_op (#13718)
16311dcd01960c33f52afb782b3d850fc2149a08cc6bc10ed6cbcc3904b7d6bbcd137589eebd899aa38fJeff Bolzjbolz@nvidia.com2025-05-23T00:45:02-04:00GitHubnoreply@github.com2025-05-23T06:45:02+02:00vulkan: support CPY from any type to itself (#13695)
1632c10ed6cbcc3904b7d6bbcd137589eebd899aa38fa127ff17800452075bb323277e3b9f8db8612cedJeff Bolzjbolz@nvidia.com2025-05-23T00:33:45-04:00GitHubnoreply@github.com2025-05-23T06:33:45+02:00vulkan: Disable coopmat/coopmat2/bfloat extensions if glslc doesn't support it (#13696)
1633a127ff17800452075bb323277e3b9f8db8612ced3079e9ac8e04ef6eddeb0c164d72edb6b6fd2df5Judd4046440+foldl@users.noreply.github.com2025-05-23T12:33:08+08:00GitHubnoreply@github.com2025-05-23T06:33:08+02:00use LOG_WARN to replace `std::cerr` (#13657)
16343079e9ac8e04ef6eddeb0c164d72edb6b6fd2df58a1d206f1d2b4e45918b589f3165b4be232f7ba8Diego Devesaslarengh@gmail.com2025-05-22T15:21:37-07:00GitHubnoreply@github.com2025-05-23T00:21:37+02:00release : fix windows hip release (#13707)
16358a1d206f1d2b4e45918b589f3165b4be232f7ba8797990c4bca0dca5be295c63e3fb2800dc0a69c2Georgi Gerganovggerganov@gmail.com2025-05-22T22:21:07+03:00GitHubnoreply@github.com2025-05-22T22:21:07+03:00tts : fix n_ubatch + make WavTokenizer cache-less (#13713)
1636797990c4bca0dca5be295c63e3fb2800dc0a69c2ab86335760ebb441574eb47f886fa1ee302e2131Xuan-Son Nguyenson@huggingface.co2025-05-22T20:42:48+02:00GitHubnoreply@github.com2025-05-22T20:42:48+02:00mtmd : add ultravox audio input (#13623)
1637ab86335760ebb441574eb47f886fa1ee302e2131cc74d5be990e37f201591fd868a92e64abdbf902Aaron Teoaaron.teo1@ibm.com2025-05-23T02:31:29+08:00GitHubnoreply@github.com2025-05-22T21:31:29+03:00common: Include torch package for s390x (#13699)
1638cc74d5be990e37f201591fd868a92e64abdbf9025be24af73d77ea23d399726f1d2a01a70ee86331Georgi Gerganovggerganov@gmail.com2025-05-22T16:33:39+03:00GitHubnoreply@github.com2025-05-22T16:33:39+03:00server : pad small embedding batches (#13692)
16395be24af73d77ea23d399726f1d2a01a70ee86331d394a9aedc50a13b7f6373416f7c1ccabfe79c32Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-22T14:25:05+02:00GitHubnoreply@github.com2025-05-22T14:25:05+02:00gguf-py : correct charsmap parameter typing (#13701)
1640d394a9aedc50a13b7f6373416f7c1ccabfe79c326b56a64690a318fcabcd7739ac7e314d44785ea8Nicolò Scipionenicolo.scipione@codeplay.com2025-05-22T13:54:43+02:00GitHubnoreply@github.com2025-05-22T12:54:43+01:00sycl : Remove waits from function calls (#13702)
16416b56a64690a318fcabcd7739ac7e314d44785ea8a4e8912dfd4604be1e39bc86ba4c0b02969967efEwan Crawfordewan@codeplay.com2025-05-22T09:24:09+01:00GitHubnoreply@github.com2025-05-22T16:24:09+08:00SYCL: Avoid using with SYCL-Graph for unsupported nodes (#13587)
1642a4e8912dfd4604be1e39bc86ba4c0b02969967efedbf42edfdabb9cea72ae12137570cf48f5d8076Henry Linjamäkihenry.mikael.linjamaki@intel.com2025-05-22T02:21:45+03:00GitHubnoreply@github.com2025-05-21T16:21:45-07:00opencl: Add support for multiple devices (#12622)
1643edbf42edfdabb9cea72ae12137570cf48f5d8076d643bb2c798df9c2cd61067d2692b1cd417df402Henry Linjamäkihenry.mikael.linjamaki@intel.com2025-05-21T23:21:17+03:00GitHubnoreply@github.com2025-05-21T13:21:17-07:00opencl: fix couple crashes (#12795)
1644d643bb2c798df9c2cd61067d2692b1cd417df4028e186ef0e764c7a620e402d1f76ebad60bf31c49Diego Devesaslarengh@gmail.com2025-05-21T13:09:57-07:00GitHubnoreply@github.com2025-05-21T22:09:57+02:00releases : build CPU backend separately (windows) (#13642)
16458e186ef0e764c7a620e402d1f76ebad60bf31c495fbfe384d4659f81c47a477eb8ee97692c7ffef9Georgi Gerganovggerganov@gmail.com2025-05-21T20:00:49+03:00GitHubnoreply@github.com2025-05-21T20:00:49+03:00hparams : support models for which all layers use SWA (#13682)
16465fbfe384d4659f81c47a477eb8ee97692c7ffef9c76532e7ba128bb097bf6836bf0f5592e1b56b76Georgi Gerganovggerganov@gmail.com2025-05-21T19:46:56+03:00GitHubnoreply@github.com2025-05-21T19:46:56+03:00server : improve error reporting (#13680)
1647c76532e7ba128bb097bf6836bf0f5592e1b56b762aa777d86d3f7bb80b93e226f1c25e47825f6a83antichristHater142441588+antichristHater@users.noreply.github.com2025-05-21T19:40:35+03:00GitHubnoreply@github.com2025-05-21T18:40:35+02:00convert : add qwen2vl support for unsloth merges (#13686)
16482aa777d86d3f7bb80b93e226f1c25e47825f6a83eb0f5c28d37126baa756117d5bdaadc62e03344eSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-21T16:57:38+02:00GitHubnoreply@github.com2025-05-21T16:57:38+02:00examples : switch retrieval to llama_encode (#13685)
1649eb0f5c28d37126baa756117d5bdaadc62e03344ecf4cb59e64d72a1b4c781f71a74de5756a4e2376Emmanuel Ferdmanemmanuelferdman@gmail.com2025-05-21T17:33:54+03:00GitHubnoreply@github.com2025-05-21T16:33:54+02:00gguf-py : display the invalid gguf type (#13687)
1650cf4cb59e64d72a1b4c781f71a74de5756a4e23760d5c74216170ef97e5e7511563837263f2d1a496Xuan-Son Nguyenson@huggingface.co2025-05-21T16:26:33+02:00GitHubnoreply@github.com2025-05-21T16:26:33+02:00ggml : add ggml_gelu_erf() (#13667)
16510d5c74216170ef97e5e7511563837263f2d1a49642158ae2e8ead667a83f07247321ce85f32ace66Robin Davidsson40024429+R-Dson@users.noreply.github.com2025-05-21T15:15:27+02:00GitHubnoreply@github.com2025-05-21T15:15:27+02:00server : Add the endpoints /api/tags and /api/chat (#13659)
165242158ae2e8ead667a83f07247321ce85f32ace66797f2ac0625b22edeff03cc30e0f988da6b6b068Dorin-Andrei Gemandoringeman@gmail.com2025-05-21T16:07:57+03:00GitHubnoreply@github.com2025-05-21T15:07:57+02:00server : fix first message identification (#13634)
1653797f2ac0625b22edeff03cc30e0f988da6b6b068b44890df2e4fad0ece1d5366dcbc8bedae23b658Georgi Gerganovggerganov@gmail.com2025-05-21T15:11:13+03:00GitHubnoreply@github.com2025-05-21T15:11:13+03:00kv-cache : simplify the interface (#13660)
1654b44890df2e4fad0ece1d5366dcbc8bedae23b65833983057d0f578aca74ba15eccc3de9c267a5ff6Georgi Gerganovggerganov@gmail.com2025-05-21T13:09:21+03:00GitHubnoreply@github.com2025-05-21T13:09:21+03:00model : disable SWA for Phi models (#13676)
165533983057d0f578aca74ba15eccc3de9c267a5ff6fb1cab201c6c4cd36731f100df74eece2f4706faR0CKSTARyeahdongcn@gmail.com2025-05-21T09:58:49+08:00GitHubnoreply@github.com2025-05-21T09:58:49+08:00musa: Upgrade MUSA SDK version to rc4.0.1 and use mudnn::Unary::IDENTITY op to accelerate D2D memory copy (#13647)
1656fb1cab201c6c4cd36731f100df74eece2f4706fab7a17463ec190aeee7b9077c606c910fb4688b84Eve139727413+netrunnereve@users.noreply.github.com2025-05-20T21:35:16ZGitHubnoreply@github.com2025-05-20T21:35:16Zvulkan: fix warnings (#13626)
1657b7a17463ec190aeee7b9077c606c910fb4688b84be0239693c1530a18496086331fc18d8a9adbad1l3utterflygc.pthzfoldr@gmail.com2025-05-21T00:55:30+08:00GitHubnoreply@github.com2025-05-20T18:55:30+02:00mtmd-helper : bug fix to token batching in mtmd (#13650)
1658be0239693c1530a18496086331fc18d8a9adbad1a4090d1174aed22dde5cacce2a4c27656b987a2fGeorgi Gerganovggerganov@gmail.com2025-05-20T19:21:04+03:00GitHubnoreply@github.com2025-05-20T19:21:04+03:00model : fix llama4 graph (#13663)
1659a4090d1174aed22dde5cacce2a4c27656b987a2fb69f1647f9953cb3773266d2c83a92fd0e7e6d66Georgi Gerganovggerganov@gmail.com2025-05-20T16:13:16+03:00GitHubnoreply@github.com2025-05-20T16:13:16+03:00llama : remove llama_kv_cache_view API + remove deprecated (#13653)
1660b69f1647f9953cb3773266d2c83a92fd0e7e6d66759e37b0d89bc4bd1bce860dc5f3c3052e08575cJohannes Gäßlerjohannesg@5d6.de2025-05-20T14:45:07+02:00GitHubnoreply@github.com2025-05-20T14:45:07+02:00CUDA: skip fully masked-out KV in FA vec kernel (#13584)
1661759e37b0d89bc4bd1bce860dc5f3c3052e08575c4245e622e0cc3af1ca3056104e465dc4d303bd7dSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-20T12:03:17+02:00GitHubnoreply@github.com2025-05-20T12:03:17+02:00tests : avoid github urls due to throttling (#13654)
16624245e622e0cc3af1ca3056104e465dc4d303bd7dc9c64dee572c5eedf073a6c6eb5d92d8283f7639Svetlozar Georgiev55534064+sgeor255@users.noreply.github.com2025-05-20T10:34:15+01:00GitHubnoreply@github.com2025-05-20T11:34:15+02:00sycl: disable reorder for sycl mulmat (#13536)
1663c9c64dee572c5eedf073a6c6eb5d92d8283f7639c00a2634bec49805c6b31438b1a6006a2bd793cb0cc4mpicard12@live.de2025-05-20T10:11:56+02:00GitHubnoreply@github.com2025-05-20T10:11:56+02:00Set GLM4 blk.*.attn_output.weight, kqv_out-* matmul to GGML_PREC_F32 to fix infinity values in output (#13639)
1664c00a2634bec49805c6b31438b1a6006a2bd793cbe298d2fbd082a52c0f6ed02729f94e9bf630cf17Georgi Gerganovggerganov@gmail.com2025-05-20T10:41:40+03:00GitHubnoreply@github.com2025-05-20T10:41:40+03:00metal : fix typo in FA kernel comments (#13651)
1665e298d2fbd082a52c0f6ed02729f94e9bf630cf17f0adb80bf7c2c0d80abb04f4533b5513622d9964Georgi Gerganovggerganov@gmail.com2025-05-20T08:05:46+03:00GitHubnoreply@github.com2025-05-20T08:05:46+03:00kv-cache : add SWA support (#13194)
1666f0adb80bf7c2c0d80abb04f4533b5513622d9964f7c9429c85748cde9599499601ba48d0057722e6Xinpeng Dou15529241576@163.com2025-05-20T11:43:43+08:00GitHubnoreply@github.com2025-05-20T11:43:43+08:00CANN: Update CANN model support (#13162)
1667f7c9429c85748cde9599499601ba48d0057722e61dfbf2cf3a9f15193dd893396d07762bbd2c4785Nicolò Scipionenicolo.scipione@codeplay.com2025-05-20T02:54:43+02:00GitHubnoreply@github.com2025-05-20T08:54:43+08:00sycl : Overcoming workaround for mmap() allocation on Windows (#13482)
16681dfbf2cf3a9f15193dd893396d07762bbd2c47858960efd0a65e5a4830a14f6937c10703534f2569psocolovsky50770545+psocolovsky@users.noreply.github.com2025-05-19T21:17:36+02:00GitHubnoreply@github.com2025-05-19T21:17:36+02:00common : add load_progress_callback (#13617)
16698960efd0a65e5a4830a14f6937c10703534f2569725f23f1f3f0d3adf49f95d8dfa6e7c74adff1490cc4mpicard12@live.de2025-05-19T17:54:08+02:00GitHubnoreply@github.com2025-05-19T17:54:08+02:00Vulkan: Add f32 accumulator support to quantized mul mat to fix GLM4 32B incoherence (#13607)
1670725f23f1f3f0d3adf49f95d8dfa6e7c74adff14992ecdcc06a4c405a415bcaa0cb772bc560aa23b1Alberto Cabrera Pérezalberto.cabrera@codeplay.com2025-05-19T14:38:20+01:00GitHubnoreply@github.com2025-05-19T14:38:20+01:00sycl : backend documentation review (#13544)
167192ecdcc06a4c405a415bcaa0cb772bc560aa23b1f71f40a2847d4c9f57b86cd206e0a27b2bfb6d1cXuan-Son Nguyenson@huggingface.co2025-05-19T13:04:14+02:00GitHubnoreply@github.com2025-05-19T13:04:14+02:00mtmd : add vision support for llama 4 (#13282)
1672f71f40a2847d4c9f57b86cd206e0a27b2bfb6d1cd30cb5a7fa17362c47e94a023276f169916e0d03Alberto Cabrera Pérezalberto.cabrera@codeplay.com2025-05-19T11:46:09+01:00GitHubnoreply@github.com2025-05-19T11:46:09+01:00ci : upgraded oneAPI version in SYCL workflows and dockerfile (#13532)
1673d30cb5a7fa17362c47e94a023276f169916e0d036c35981a643d4f74a286268b62f65bfa156af2e6Georgi Gerganovggerganov@gmail.com2025-05-19T12:50:29+03:00Georgi Gerganovggerganov@gmail.com2025-05-19T13:29:56+03:00sync : ggml
16746c35981a643d4f74a286268b62f65bfa156af2e68b5e19aea6ce9fe4452598663924373234041440Johannes Gäßlerjohannesg@5d6.de2025-05-19T09:33:35+02:00Georgi Gerganovggerganov@gmail.com2025-05-19T13:29:56+03:00mnist: fix segmentation fault (ggml/1227)
16758b5e19aea6ce9fe445259866392437323404144060aea028b575ab54e0dfbb31db641bb93d2ae8c4Diego Devesaslarengh@gmail.com2025-05-18T18:30:13-07:00Georgi Gerganovggerganov@gmail.com2025-05-19T13:29:56+03:00ggml : fix apple OS check in ggml_print_backtrace (ggml/1229)
167660aea028b575ab54e0dfbb31db641bb93d2ae8c49c55e5c5c24990dd17fd6c8f4f2159052d2b06f1Daniel Tangdanielzgtg.opensource@gmail.com2025-05-17T19:06:26-04:00Georgi Gerganovggerganov@gmail.com2025-05-19T13:29:56+03:00ggml : Fix missing backtrace on Linux (ggml/1228)
16779c55e5c5c24990dd17fd6c8f4f2159052d2b06f133d7aed4a83e7bbecac4af535208d4ed3e1ca8fdNick0x0b4ac@gmail.com2025-05-19T18:25:41+08:00GitHubnoreply@github.com2025-05-19T13:25:41+03:00fix: check model pointer validity before use (#13631)
167833d7aed4a83e7bbecac4af535208d4ed3e1ca8fd6a2bc8bfb7cd502e5ebc72e36c97a6f848c21c2cChenguang Li757486878@qq.com2025-05-19T14:21:17+08:00GitHubnoreply@github.com2025-05-19T14:21:17+08:00CANN: Support MOE Model MUL_MAT_ID (#13042)
16796a2bc8bfb7cd502e5ebc72e36c97a6f848c21c2ce3a7cf6c5bf6a0a24217f88607b06e4405a2b5d9Isaac McFadyenisaac@imcf.me2025-05-17T17:59:48-04:00GitHubnoreply@github.com2025-05-17T23:59:48+02:00server : added --no-prefill-assistant flag (#13608)
1680e3a7cf6c5bf6a0a24217f88607b06e4405a2b5d9518329b2d4434d0683c30b741b4f4cf5734b5f99Gilad S.7817232+giladgd@users.noreply.github.com2025-05-17T21:26:43+03:00GitHubnoreply@github.com2025-05-17T15:26:43-03:00cmake: use the current build config for vulkan-shaders-gen (#13595)
1681518329b2d4434d0683c30b741b4f4cf5734b5f992f5a4e1e09567e09be8b84a5f976334de9539d17Georgi Gerganovggerganov@gmail.com2025-05-17T12:58:55+03:00GitHubnoreply@github.com2025-05-17T12:58:55+03:00parallel : add option for non-shared and larger prompts (#13598)
16822f5a4e1e09567e09be8b84a5f976334de9539d174f41ee11d6a4ddb02e4644922bf0b56880ee6f55Jeff Bolzjbolz@nvidia.com2025-05-17T16:14:55+09:00GitHubnoreply@github.com2025-05-17T09:14:55+02:00vulkan: move common FA code to flash_attn_base.comp (#13556)
16834f41ee11d6a4ddb02e4644922bf0b56880ee6f553e0be1cacef290c99cbb99ceaa433b4344f87355Jeff Bolzjbolz@nvidia.com2025-05-17T15:35:47+09:00GitHubnoreply@github.com2025-05-17T08:35:47+02:00vulkan: use scalar FA rather than coopmat2 when N==1 (#13554)
16843e0be1cacef290c99cbb99ceaa433b4344f873556aa892ec2aa7fe0c93e87c4b970d83a942fb9454Zcoffeevampirebusiness@gmail.com2025-05-16T14:56:28-06:00GitHubnoreply@github.com2025-05-16T22:56:28+02:00llguidance : official v0.7.20 release (no actual changes) [noci] (#13594)
16856aa892ec2aa7fe0c93e87c4b970d83a942fb9454aea9f8b4e73876739bf88fb705502f291294e469Xuan-Son Nguyenson@huggingface.co2025-05-16T21:50:00+02:00GitHubnoreply@github.com2025-05-16T21:50:00+02:00server : do not return error out of context (with ctx shift disabled) (#13577)
1686aea9f8b4e73876739bf88fb705502f291294e46906c1e4abc1e50ef6dcf6369f9685ca8fe82d21feXuan-Son Nguyenson@huggingface.co2025-05-16T21:49:01+02:00GitHubnoreply@github.com2025-05-16T21:49:01+02:00webui : improve accessibility for visually impaired people (#13551)
168706c1e4abc1e50ef6dcf6369f9685ca8fe82d21fe415e40a357002d5d9b7a97ef20ccea9f4ae04c80Xuan-Son Nguyenson@huggingface.co2025-05-16T20:04:18+02:00GitHubnoreply@github.com2025-05-16T20:04:18+02:00readme : add list of dependencies and their license (#13591)
1688415e40a357002d5d9b7a97ef20ccea9f4ae04c80654a67794f7a420c6931de2f4c145eaaade5dd16Diego Devesaslarengh@gmail.com2025-05-16T10:36:51-07:00GitHubnoreply@github.com2025-05-16T19:36:51+02:00releases : use arm version of curl for arm releases (#13592)
1689654a67794f7a420c6931de2f4c145eaaade5dd165364ae4ba53cc6367b8c8bf78876839122ca4e57Georgi Gerganovggerganov@gmail.com2025-05-16T20:32:58+03:00GitHubnoreply@github.com2025-05-16T20:32:58+03:00metal : add FA-vec kernel for head size 64 (#13583)
16905364ae4ba53cc6367b8c8bf78876839122ca4e577c07ac244d59c833bf209582f6df019a77cdda59Diego Devesaslarengh@gmail.com2025-05-16T07:38:07-07:00GitHubnoreply@github.com2025-05-16T16:38:07+02:00llama : print hint when loading a model when no backends are loaded (#13589)
16917c07ac244d59c833bf209582f6df019a77cdda590a338ed013c23aecdce6449af736a35a465fa60fSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-16T14:54:23+02:00GitHubnoreply@github.com2025-05-16T14:54:23+02:00ci : add ppc64el to build-linux-cross (#13575)
16920a338ed013c23aecdce6449af736a35a465fa60fbc098c3cf0aac93e57e9bda9d95e2456acf88894Łukasz Ślusarczyk112692748+lslusarczyk@users.noreply.github.com2025-05-16T12:15:29+02:00GitHubnoreply@github.com2025-05-16T18:15:29+08:00sycl : fixed compilation warnings (#13582)
1693bc098c3cf0aac93e57e9bda9d95e2456acf88894c6a2c9e7411f54b0770b319740561bbd6a2ebd27Olivier Chafikolivier.chafik@gmail.com2025-05-15T23:29:10+01:00GitHubnoreply@github.com2025-05-15T23:29:10+01:00minja: sync (qwen3) (#13573)
1694c6a2c9e7411f54b0770b319740561bbd6a2ebd2707ad2b6db3c117868728e2cdfe3b711473d66e14Diego Devesaslarengh@gmail.com2025-05-15T10:13:11-07:00GitHubnoreply@github.com2025-05-15T19:13:11+02:00gguf : use ggml log system (#13571)
169507ad2b6db3c117868728e2cdfe3b711473d66e14c531edfa34fec8074d0b424396cdd450df23b612Daniel Tangdanielzgtg.opensource@gmail.com2025-05-15T12:47:10-04:00GitHubnoreply@github.com2025-05-15T18:47:10+02:00gguf-py : fix disconnect-before-connect in editor-gui (#13569)
1696c531edfa34fec8074d0b424396cdd450df23b61202cdd2d8b092b5a4bb18e013c6887ce49ba20ac5Xuan-Son Nguyenson@huggingface.co2025-05-15T17:40:07+02:00GitHubnoreply@github.com2025-05-15T17:40:07+02:00convert : fix conversion for llama 4 (#13567)
169702cdd2d8b092b5a4bb18e013c6887ce49ba20ac564bb51cf90d3eede8c150a23d59a0c718b78065bAtharva Dubeyatharva.dubey@codeplay.com2025-05-15T16:39:52+01:00GitHubnoreply@github.com2025-05-15T17:39:52+02:00sycl: simplify bin_bcast_kernel (#13383)
169864bb51cf90d3eede8c150a23d59a0c718b78065b9c404ed54c3c8d8d2aa3153313766c8286739387Svetlozar Georgiev55534064+sgeor255@users.noreply.github.com2025-05-15T16:35:44+01:00GitHubnoreply@github.com2025-05-15T17:35:44+02:00sycl: reordered Q4_K MMVQ (#13109)
16999c404ed54c3c8d8d2aa3153313766c82867393876c8b91500e75df6664278d1e9af3e39e8a2fb0d0Łukasz Ślusarczyk112692748+lslusarczyk@users.noreply.github.com2025-05-15T16:53:41+02:00GitHubnoreply@github.com2025-05-15T16:53:41+02:00sycl: use oneDNN for matrices multiplication (#12972)
17006c8b91500e75df6664278d1e9af3e39e8a2fb0d03cc1f1f1d24472a6558c942b1c78989ff4b0e569Diego Devesaslarengh@gmail.com2025-05-15T06:46:55-07:00GitHubnoreply@github.com2025-05-15T15:46:55+02:00llama-bench : fix -ot with dl backends (#13563)
17013cc1f1f1d24472a6558c942b1c78989ff4b0e569c753d7bed0dc2cc2d5c42dfa9806cba91748392eXuan-Son Nguyenson@huggingface.co2025-05-15T14:24:50+02:00GitHubnoreply@github.com2025-05-15T14:24:50+02:00webui : handle PDF input (as text or image) + convert pasted long content to file (#13562)
1702c753d7bed0dc2cc2d5c42dfa9806cba91748392eb2838049ccf50859cea4c390e81405c2fb01e820Piotr Wilkin (ilintar)piotr.wilkin@syndatis.com2025-05-15T08:40:58+02:00GitHubnoreply@github.com2025-05-15T08:40:58+02:00server : proper error handling for missing elements in messages array (OpenAI compatible backend) (#13540)
1703b2838049ccf50859cea4c390e81405c2fb01e820aa48e373f256df9608395ee6881e7010840d6202Georgi Gerganovggerganov@gmail.com2025-05-15T05:57:02+03:00GitHubnoreply@github.com2025-05-15T05:57:02+03:00bench : handle decode errors (#13548)
1704aa48e373f256df9608395ee6881e7010840d6202e3a9421b78da5c810d812e6348a405f5acc39f34Olivier Chafikochafik@users.noreply.github.com2025-05-15T02:39:51+01:00GitHubnoreply@github.com2025-05-15T02:39:51+01:00`server`: inject date_string in llama 3.x template + fix date for firefunction v2 (#12802)
1705e3a9421b78da5c810d812e6348a405f5acc39f345ab5d5fb256aa23f8ab4de8463515ea627f43006Georgi Gerganovggerganov@gmail.com2025-05-14T23:15:15+03:00GitHubnoreply@github.com2025-05-14T23:15:15+03:00kv-cache : fix out-of-bounds view during reserve graph (#13547)
17065ab5d5fb256aa23f8ab4de8463515ea627f430063198405e98530c683d12fd123e3920f2bd2aafa5Yibo Caicyb70289@gmail.com2025-05-15T03:53:52+08:00GitHubnoreply@github.com2025-05-14T21:53:52+02:00arm64: optimize q6_k_q8_k kernel with i8mm (#13519)
17073198405e98530c683d12fd123e3920f2bd2aafa5f5170c1d7a66222ca7c75d2022fec3ed87257e0bOlivier Chafikochafik@users.noreply.github.com2025-05-14T19:50:57+01:00GitHubnoreply@github.com2025-05-14T19:50:57+01:00`common`: add partial regex support (#12808)
1708f5170c1d7a66222ca7c75d2022fec3ed87257e0b017f10b5fa630a013ec4f9936e410a60d4f460d5Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-14T20:22:49+02:00GitHubnoreply@github.com2025-05-14T21:22:49+03:00editorconfig : fix trailing whitespace from #13542 (#13546)
1709017f10b5fa630a013ec4f9936e410a60d4f460d54696d5674999dc10a7fb8c27b33406a929f7463aGilad S.7817232+giladgd@users.noreply.github.com2025-05-14T19:18:18+03:00GitHubnoreply@github.com2025-05-14T19:18:18+03:00fix: crash when calling `llama_state_get_size` on a context without a KV cache (#13542)
17104696d5674999dc10a7fb8c27b33406a929f7463ab7d26720821823e23e2273a99e38398d511242e9Johannes Gäßlerjohannesg@5d6.de2025-05-14T16:41:02+02:00GitHubnoreply@github.com2025-05-14T16:41:02+02:00CUDA: fix crash on large batch size for quant. MoE (#13537)
1711b7d26720821823e23e2273a99e38398d511242e96da34fa27620fa56e3334172e023f4f2533df51fDiego Devesaslarengh@gmail.com2025-05-14T07:12:36-07:00GitHubnoreply@github.com2025-05-14T16:12:36+02:00llama : fix quantize with dl backends (#13539)
17126da34fa27620fa56e3334172e023f4f2533df51f5e7d95e22e386d316f7f659b74c9c34b65507912Johannes Gäßlerjohannesg@5d6.de2025-05-14T16:08:20+02:00GitHubnoreply@github.com2025-05-14T16:08:20+02:00CUDA: faster Deepseek FA, add Turing support (#13435)
17135e7d95e22e386d316f7f659b74c9c34b65507912053174436f3b3cbb01077f26ff17809537b832c7Gabe Goodhartghart@us.ibm.com2025-05-14T06:53:59-06:00GitHubnoreply@github.com2025-05-14T15:53:59+03:00fix: Move build_inp_pos to the top of the graph section for build_granite (#13538)
1714053174436f3b3cbb01077f26ff17809537b832c7360a9c98e13d35f322b4c5b1309aab0cc90ed82bGeorgi Gerganovggerganov@gmail.com2025-05-14T15:42:10+03:00GitHubnoreply@github.com2025-05-14T15:42:10+03:00server : passthrough the /models endpoint during loading (#13535)
1715360a9c98e13d35f322b4c5b1309aab0cc90ed82b09d13d94fb169093095416ac6323551c37b75339Xuan-Son Nguyenson@huggingface.co2025-05-14T13:35:07+02:00GitHubnoreply@github.com2025-05-14T13:35:07+02:00server : fix cache_tokens bug with no cache_prompt (#13533)
171609d13d94fb169093095416ac6323551c37b7533924e86cae7219b0f3ede1d5abdf5bf3ad515cccb8bandoti141645996+bandoti@users.noreply.github.com2025-05-14T07:53:57-03:00GitHubnoreply@github.com2025-05-14T07:53:57-03:00cmake: simplify vulkan shader test logic (#13263)
171724e86cae7219b0f3ede1d5abdf5bf3ad515cccb8bb1681fbd532eba26ae4c14cd8be884c8afeb31cJeff Bolzjbolz@nvidia.com2025-05-14T18:55:26+09:00GitHubnoreply@github.com2025-05-14T11:55:26+02:00vulkan: KHR_coopmat flash attention (#13506)
1718bb1681fbd532eba26ae4c14cd8be884c8afeb31cd486dd3e8ecfba0170f8632a1530540de560f4a3Xuan-Son Nguyenson@huggingface.co2025-05-14T10:26:12+02:00GitHubnoreply@github.com2025-05-14T10:26:12+02:00webui : use fflate for more deterministic gzip compress (#13525)
1719d486dd3e8ecfba0170f8632a1530540de560f4a321ca987fba504d273ea28ebc4d3e5b3736a11c8eLuca Stefaniluca.stefani.ge1@gmail.com2025-05-14T10:07:31+02:00GitHubnoreply@github.com2025-05-14T10:07:31+02:00webui: Allow pasting file from clipboard (#13526)
172021ca987fba504d273ea28ebc4d3e5b3736a11c8ebe1d4a13db26750fac702ceb3af88ae4f39dc9f4ddpasa112642920+ddpasa@users.noreply.github.com2025-05-14T09:59:12+02:00GitHubnoreply@github.com2025-05-14T09:59:12+02:00docs: Update link to ggml-org in multimodal.md (#13513)
1721be1d4a13db26750fac702ceb3af88ae4f39dc9f4ab3971f2a0a526564bfde0e4cc8a5b90f9d33ad2Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-14T08:41:01+02:00GitHubnoreply@github.com2025-05-14T08:41:01+02:00scripts : fix compare-llama-bench.py show parameter (#13514)
1722ab3971f2a0a526564bfde0e4cc8a5b90f9d33ad2e5c834f718a32b7584f142799bbf508fddb9021cJeff Bolzjbolz@nvidia.com2025-05-14T13:15:50+09:00GitHubnoreply@github.com2025-05-14T06:15:50+02:00vulkan: workaround FA compile failures on macos (#13517)
1723e5c834f718a32b7584f142799bbf508fddb9021c71bdbdb58757d508557e6d8b387f666cdfb25c5eEd Addario29247825+EAddario@users.noreply.github.com2025-05-13T18:12:31+01:00GitHubnoreply@github.com2025-05-13T19:12:31+02:00quantize : improve tensor-type pattern matching (#13033)
172471bdbdb58757d508557e6d8b387f666cdfb25c5ef0995d28ce3d15095b6845d94ce4465e46575873Xuan-Son Nguyenson@huggingface.co2025-05-13T17:07:21+02:00GitHubnoreply@github.com2025-05-13T17:07:21+02:00clip : clip.h become private API (⚠️ breaking change) (#13510)
1725f0995d28ce3d15095b6845d94ce4465e46575873c252e0c4097b34666e5a81db9d0450d71fa3098fGeorgi Gerganovggerganov@gmail.com2025-05-13T18:04:39+03:00GitHubnoreply@github.com2025-05-13T18:04:39+03:00metal : use FA-vec kernel up to batch size 20 (#13496)
1726c252e0c4097b34666e5a81db9d0450d71fa3098f4f711afed5e7ef4304b567c8888ee1aa60e868ebGeorgi Gerganovggerganov@gmail.com2025-05-13T18:04:00+03:00GitHubnoreply@github.com2025-05-13T18:04:00+03:00metal : optimize multi-sequence FA vec kernel (#13493)
17274f711afed5e7ef4304b567c8888ee1aa60e868ebb89d605a91dee0a518ecd582f8991a07d523e2faDan Johanssondan.johansson@arm.com2025-05-13T17:02:28+02:00GitHubnoreply@github.com2025-05-13T18:02:28+03:00ggml-cpu: Update KleidiAI to v1.6 and fix include directives (#13509)
1728b89d605a91dee0a518ecd582f8991a07d523e2fab4726345aca49e2ad62d615e6e370b3dbad6434fGeorgi Gerganovggerganov@gmail.com2025-05-13T18:01:53+03:00GitHubnoreply@github.com2025-05-13T18:01:53+03:00batched-bench : fix pp batch contents (#13492)
1729b4726345aca49e2ad62d615e6e370b3dbad6434fbf7937112058f2815fc3825a9ff7b536ecafa3bbXuan-Son Nguyenson@huggingface.co2025-05-13T15:33:58+02:00GitHubnoreply@github.com2025-05-13T15:33:58+02:00mtmd : remove libllava, remove clip-quantize-cli (⚠️ breaking change) (#13460)
1730bf7937112058f2815fc3825a9ff7b536ecafa3bbd590cd4c244e5f260c42c290b83a358b9d86d763Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-13T15:31:12+02:00GitHubnoreply@github.com2025-05-13T15:31:12+02:00scripts : support arbitrary input file formats in compare-llama-bench.py (#13455)
1731d590cd4c244e5f260c42c290b83a358b9d86d7631e2809bc4b5d8db2a9ed12ac872eca832c53f5fdGabe Goodhartghart@us.ibm.com2025-05-13T07:12:01-06:00GitHubnoreply@github.com2025-05-13T15:12:01+02:00model : Granite MoE shared (#13269)
17321e2809bc4b5d8db2a9ed12ac872eca832c53f5fdcf0a43bb6490bd49344775abb22ba26f8047cb54Georgi Gerganovggerganov@gmail.com2025-05-13T14:01:45+03:00Georgi Gerganovggerganov@gmail.com2025-05-13T14:02:28+03:00sync : ggml
1733cf0a43bb6490bd49344775abb22ba26f8047cb54f0d46ef15717cd609a7b69cf6190edde64d466c8Diego Devesaslarengh@gmail.com2025-05-12T15:31:37-07:00GitHubnoreply@github.com2025-05-13T00:31:37+02:00llama-bench : add defrag-thold, check for invalid ranges (#13487)
1734f0d46ef15717cd609a7b69cf6190edde64d466c8de4c07f93783a1a96456a44dc16b9db538ee1618lhezquic_lih@quicinc.com2025-05-12T13:13:49-07:00GitHubnoreply@github.com2025-05-12T13:13:49-07:00opencl: remove unnecessary assert for `add` (#13257)
1735de4c07f93783a1a96456a44dc16b9db538ee161810d2af0eaa0aafd7c6577b279dfa5221ff44a63fXuan-Son Nguyenson@huggingface.co2025-05-12T15:06:51+02:00GitHubnoreply@github.com2025-05-12T15:06:51+02:00clip : cap max image size 1024 for qwen vl model (#13478)
173610d2af0eaa0aafd7c6577b279dfa5221ff44a63f064cc596ac44308dc326a17c9e3163c34a6f29d1Johannes Gäßlerjohannesg@5d6.de2025-05-12T14:44:49+02:00GitHubnoreply@github.com2025-05-12T14:44:49+02:00llama/ggml: add LLM training support (#10544)
1737064cc596ac44308dc326a17c9e3163c34a6f29d191159ee9df84edb72ccf874e353d2414f3a8c60dGeorgi Gerganovggerganov@gmail.com2025-05-12T15:12:27+03:00GitHubnoreply@github.com2025-05-12T15:12:27+03:00context : fix state io for memory-less contexts (#13470)
173891159ee9df84edb72ccf874e353d2414f3a8c60d22cdab343b63edd0906f1c132616746085f81983Anudit Nagarnagaranudit@gmail.com2025-05-12T18:56:42+07:00GitHubnoreply@github.com2025-05-12T13:56:42+02:00server : allow content to be null in oaicompat_completion_params_parse (#13477)
173922cdab343b63edd0906f1c132616746085f81983a71a4075cdb8e81eaaa834e48ffda88b038286bcDiego Devesaslarengh@gmail.com2025-05-12T13:08:22+02:00GitHubnoreply@github.com2025-05-12T13:08:22+02:00llama-bench : accept ranges for integer parameters (#13410)
1740a71a4075cdb8e81eaaa834e48ffda88b038286bc95e18884fc7ea4031f70f1a518d5d1df616e5717Dan Johanssondan.johansson@arm.com2025-05-12T13:06:19+02:00GitHubnoreply@github.com2025-05-12T13:06:19+02:00ggml-cpu: Integrate fp32=bf16xbf16 SME KleidiAI kernel (#13053)
174195e18884fc7ea4031f70f1a518d5d1df616e5717df8491922f0ea4e032338186350dff2fb6b2b4e3Johannes Gäßlerjohannesg@5d6.de2025-05-12T10:51:21+02:00GitHubnoreply@github.com2025-05-12T10:51:21+02:00CUDA: fix misaligned synchronization in FA (#13469)
1742df8491922f0ea4e032338186350dff2fb6b2b4e314492144c286bbf38bb1903128403d9e2ebad54cXuan-Son Nguyenson@huggingface.co2025-05-12T10:29:13+02:00GitHubnoreply@github.com2025-05-12T10:29:13+02:00ggml : add mrope kernel for metal (#13457)
174314492144c286bbf38bb1903128403d9e2ebad54cc104023994d36a8e791fc6a43789b84fd552cefcAtharva Dubeyatharva.dubey@codeplay.com2025-05-12T06:15:32+01:00GitHubnoreply@github.com2025-05-12T13:15:32+08:00enable dpcpp nightly builds with libraries (#13406)
1744c104023994d36a8e791fc6a43789b84fd552cefc9a390c4829cd3058d26a2e2c09d16e3fd12bf1b1City125218114+city96@users.noreply.github.com2025-05-12T00:39:06+02:00GitHubnoreply@github.com2025-05-12T00:39:06+02:00mtmd : Use RMS norm for InternVL 3 38B and 78B mmproj (#13459)
17459a390c4829cd3058d26a2e2c09d16e3fd12bf1b109232370fc6426aa5dd9be01a8271b9c28f5af3aAnthony Umferaumfer@gmail.com2025-05-11T11:08:26-04:00GitHubnoreply@github.com2025-05-11T17:08:26+02:00tools : fix uninitialized llama_batch in server (#13436)
174609232370fc6426aa5dd9be01a8271b9c28f5af3a7474e00b34629e9cd8b06bc87ad935584ea30f8eSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-11T16:20:39+02:00GitHubnoreply@github.com2025-05-11T16:20:39+02:00scripts : exit compare-llama-bench.py gracefully when there's nothing to compare (#13451)
17477474e00b34629e9cd8b06bc87ad935584ea30f8e7f323a589f8684c0eb722e7309074cb5eac0c8b5Johannes Gäßlerjohannesg@5d6.de2025-05-11T16:09:33+02:00GitHubnoreply@github.com2025-05-11T16:09:33+02:00CUDA: fix crash with partial offloading of MoE (#13439)
17487f323a589f8684c0eb722e7309074cb5eac0c8b53eac209319a6726fd9687c6188fc6b916b65953dDavid Huang1969802+hjc4869@users.noreply.github.com2025-05-11T20:18:39+08:00GitHubnoreply@github.com2025-05-11T14:18:39+02:00Add `--no-op-offload` to improve `-ot` pp perf in MoE models like llama4 400B (#13386)
17493eac209319a6726fd9687c6188fc6b916b65953da634d75d1bb4034b8c945042ceea268b5b895730City125218114+city96@users.noreply.github.com2025-05-11T11:35:52+02:00GitHubnoreply@github.com2025-05-11T11:35:52+02:00mtmd : support InternVL 3 38B and 78B mmproj (#13443)
1750a634d75d1bb4034b8c945042ceea268b5b89573062d4250e52917dc4d634f054b1e2183ed7dee944Xuan-Son Nguyenson@huggingface.co2025-05-11T11:34:23+02:00GitHubnoreply@github.com2025-05-11T11:34:23+02:00mtmd : move helpers to dedicated file (#13442)
175162d4250e52917dc4d634f054b1e2183ed7dee9440208355f42bdab88a08507ead4a6302790a08323Thomas Germer99991@users.noreply.github.com2025-05-10T22:26:46+02:00GitHubnoreply@github.com2025-05-10T22:26:46+02:00docs : Fix typo in InternVL3 model name (#13440)
17520208355f42bdab88a08507ead4a6302790a08323d2a4ef05c60506ee48e7375eb36f2257de7ab0d2Johannes Gäßlerjohannesg@5d6.de2025-05-10T22:22:48+02:00GitHubnoreply@github.com2025-05-10T22:22:48+02:00CUDA: fix race conditions FlashAttention kernels (#13438)
1753d2a4ef05c60506ee48e7375eb36f2257de7ab0d215e6125a397f6086c1dfdf7584acdb7c730313dcSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-10T22:08:07+02:00GitHubnoreply@github.com2025-05-10T22:08:07+02:00vocab : add ByteDance-Seed/Seed-Coder (#13423)
175415e6125a397f6086c1dfdf7584acdb7c730313dc3b24d26c22b9d92b5aa39930988700c84e524cf4Xuan-Son Nguyenson@huggingface.co2025-05-10T19:57:54+02:00GitHubnoreply@github.com2025-05-10T19:57:54+02:00mtmd : add hard limit on image resolution for qwen2vl / qwen2.5vl (#13434)
17553b24d26c22b9d92b5aa39930988700c84e524cf443dfd741a5da77982e0a94d1e522a2481dfb914dXuan-Son Nguyenson@huggingface.co2025-05-10T18:44:49+02:00GitHubnoreply@github.com2025-05-10T18:44:49+02:00server : update docs (#13432)
175643dfd741a5da77982e0a94d1e522a2481dfb914db064a51a4e7246fa43a3307f58e59f65e6be170aSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-10T17:19:52+02:00GitHubnoreply@github.com2025-05-10T17:19:52+02:00llguidance : set tokenizer slices to default (#13424)
1757b064a51a4e7246fa43a3307f58e59f65e6be170a053367d149f778cdabc356ee3024494e0dd53223Thammachart Chinvarapon1731496+Thammachart@users.noreply.github.com2025-05-10T21:34:48+07:00GitHubnoreply@github.com2025-05-10T16:34:48+02:00ci: free_disk_space flag enabled for intel variant (#13426)
1758053367d149f778cdabc356ee3024494e0dd53223d8919424f1dee7dc1638349c616f2ef5d2ee16fbXuan-Son Nguyenson@huggingface.co2025-05-10T16:26:42+02:00GitHubnoreply@github.com2025-05-10T16:26:42+02:00mtmd : support InternVL 2.5 and 3 (#13422)
1759d8919424f1dee7dc1638349c616f2ef5d2ee16fb7fef11766cdeb9fa7bbbe3db13580616b7d3d599Johannes Gäßlerjohannesg@5d6.de2025-05-10T09:16:52+02:00GitHubnoreply@github.com2025-05-10T09:16:52+02:00CUDA: fix FlashAttention on Turing (#13415)
17607fef11766cdeb9fa7bbbe3db13580616b7d3d599dc1d2adfc0f4de84da7923866d00781ec5c4e666Xuan-Son Nguyenson@huggingface.co2025-05-10T08:16:29+02:00GitHubnoreply@github.com2025-05-10T08:16:29+02:00arg : add env var to control mmproj (#13416)
1761dc1d2adfc0f4de84da7923866d00781ec5c4e6667c28a74e0783f4bb74a246fb9f19bf212139e365Jeff Bolzjbolz@nvidia.com2025-05-09T23:07:07-07:00GitHubnoreply@github.com2025-05-10T08:07:07+02:00vulkan: scalar flash attention implementation (#13324)
17627c28a74e0783f4bb74a246fb9f19bf212139e36533eff4024084d1f0c8441b79f7208a52fad79858Helton Reis47722840+HRKings@users.noreply.github.com2025-05-09T17:15:39-03:00GitHubnoreply@github.com2025-05-09T23:15:39+03:00chore(llguidance): use tagged version that does not break the build (#13413)
176333eff4024084d1f0c8441b79f7208a52fad7985817512a94d636c4b6c1332370acb3e5af3ca70918Xuan-Son Nguyenson@huggingface.co2025-05-09T19:29:37+02:00GitHubnoreply@github.com2025-05-09T19:29:37+02:00 server : vision support via libmtmd (#12898)
176417512a94d636c4b6c1332370acb3e5af3ca70918611aa914ef4231fab5d1ad04773c42e119ae2d2eAlberto Cabrera Pérezalberto.cabrera@codeplay.com2025-05-09T16:34:08+01:00GitHubnoreply@github.com2025-05-09T16:34:08+01:00sycl : implementation of reordered Q4_0 MMVQ for Intel GPUs (#12858)
1765611aa914ef4231fab5d1ad04773c42e119ae2d2e0cf6725e9f9a164c39f7a87214d60342f7f946d8Georgi Gerganovggerganov@gmail.com2025-05-09T15:14:56+03:00GitHubnoreply@github.com2025-05-09T15:14:56+03:00metal : optimize MoE for large batches (#13388)
17660cf6725e9f9a164c39f7a87214d60342f7f946d827ebfcacbaadc6104e2b18acd8f13515cbf63dceJohannes Gäßlerjohannesg@5d6.de2025-05-09T13:34:58+02:00GitHubnoreply@github.com2025-05-09T13:34:58+02:00CUDA: FA support for Deepseek (Ampere or newer) (#13306)
176727ebfcacbaadc6104e2b18acd8f13515cbf63dce5c86c9ed3ef1cc7307fdce05f0f0e2e45253cf90Diego Devesaslarengh@gmail.com2025-05-09T13:02:07+02:00GitHubnoreply@github.com2025-05-09T13:02:07+02:00llama : do not crash if there is no CPU backend (#13395)
17685c86c9ed3ef1cc7307fdce05f0f0e2e45253cf90efb8b47eda78ea8ae570d4fece3953aae499289eJohannes Gäßlerjohannesg@5d6.de2025-05-09T12:14:04+02:00GitHubnoreply@github.com2025-05-09T12:14:04+02:00CUDA: fix crash on large batch size for MoE models (#13384)
1769efb8b47eda78ea8ae570d4fece3953aae499289e0527771dd80bd18479dfaaa0a98be297fc3592bfBartowski3266127+bartowski1182@users.noreply.github.com2025-05-09T05:53:58-04:00GitHubnoreply@github.com2025-05-09T11:53:58+02:00imatrix : Add --parse-special for enabling parsing of special tokens in imatrix calculation (#13389)
17700527771dd80bd18479dfaaa0a98be297fc3592bf2189fd3b6327a1d17893694125da8edcf74a6468R0CKSTARxiaodong.ye@mthreads.com2025-05-09T17:25:50+08:00GitHubnoreply@github.com2025-05-09T10:25:50+01:00llama-run: add support for downloading models from ModelScope (#13370)
17712189fd3b6327a1d17893694125da8edcf74a64683f96aeff394e9b72bbd2fa665c3e023a70ed8648Xuan-Son Nguyenson@huggingface.co2025-05-09T11:18:02+02:00GitHubnoreply@github.com2025-05-09T11:18:02+02:00mtmd : fix batch_view for m-rope (#13397)
17723f96aeff394e9b72bbd2fa665c3e023a70ed8648b486ba05bf973aae3652b3fd593e0b257d3a41d4Xuan-Son Nguyenson@huggingface.co2025-05-09T11:17:51+02:00GitHubnoreply@github.com2025-05-09T11:17:51+02:00llama : one-off chat template fix for Mistral-Small-2503 (#13398)
1773b486ba05bf973aae3652b3fd593e0b257d3a41d402115dcd9a6d0c03b527d5bee8c1493ab819ddbdRadoslav Gerganovrgerganov@gmail.com2025-05-09T10:31:07+03:00GitHubnoreply@github.com2025-05-09T10:31:07+03:00rpc : add rpc_msg_set_tensor_hash_req (#13353)
177402115dcd9a6d0c03b527d5bee8c1493ab819ddbdd9c4accaff30926e8a5fd8a2429e549158a845e0Jeff Bolzjbolz@nvidia.com2025-05-09T02:23:41-05:00GitHubnoreply@github.com2025-05-09T09:23:41+02:00vulkan: Allow up to 4096 elements for mul_mat_id row_ids (#13326)
1775d9c4accaff30926e8a5fd8a2429e549158a845e015e03282bb432631193464100c2237a3b6bcfe4cXuan-Son Nguyenson@huggingface.co2025-05-09T09:06:37+02:00GitHubnoreply@github.com2025-05-09T09:06:37+02:00server : (webui) rename has_multimodal --> modalities (#13393)
177615e03282bb432631193464100c2237a3b6bcfe4cf05a6d71a0f3dbf0730b56a1abbad41c0f42e63dDiego Devesaslarengh@gmail.com2025-05-08T23:45:22+02:00GitHubnoreply@github.com2025-05-08T23:45:22+02:00ci : limit write permission to only the release step + fixes (#13392)
1777f05a6d71a0f3dbf0730b56a1abbad41c0f42e63dee01d71e585f4a17ae83ec55d77acfdcf0bfa798Matt Clayton156335168+mattjcly@users.noreply.github.com2025-05-08T14:25:39-04:00GitHubnoreply@github.com2025-05-08T20:25:39+02:00mtmd : Expose helper_decode_image_chunk (#13366)
1778ee01d71e585f4a17ae83ec55d77acfdcf0bfa7988c83449cb780c201839653812681c3a4cf17feedXuan-Son Nguyenson@huggingface.co2025-05-08T18:51:45+02:00GitHubnoreply@github.com2025-05-08T18:51:45+02:00server : (webui) fix a very small misalignment (#13387)
17798c83449cb780c201839653812681c3a4cf17feed1a844be132dbe865358e1de81136920c1d35ac73Xuan-Son Nguyenson@huggingface.co2025-05-08T15:37:29+02:00GitHubnoreply@github.com2025-05-08T15:37:29+02:00server : (webui) revamp the input area, plus many small UI improvements (#13365)
17801a844be132dbe865358e1de81136920c1d35ac730ccc1213549e39ef4c1affb1bf5f49651ef4ce48Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-08T15:34:29+02:00GitHubnoreply@github.com2025-05-08T15:34:29+02:00convert : support rope_scaling type and rope_type (#13349)
17810ccc1213549e39ef4c1affb1bf5f49651ef4ce486562e5a4d6c58326dcd79002ea396d4141f1b18ewelixtaichitary@gmail.com2025-05-08T22:03:53+09:00GitHubnoreply@github.com2025-05-08T15:03:53+02:00mtmd : fix the calculation of n_tokens for smolvlm (#13381)
17826562e5a4d6c58326dcd79002ea396d4141f1b18e51fb96b1ff2e1cc98b2492a012b7d93531a6a9a8Georgi Gerganovggerganov@gmail.com2025-05-08T14:28:33+03:00GitHubnoreply@github.com2025-05-08T14:28:33+03:00context : allow cache-less context for embeddings (#13108)
178351fb96b1ff2e1cc98b2492a012b7d93531a6a9a870a6991edf1b60e7afa8962f830320583f3babb0Georgi Gerganovggerganov@gmail.com2025-05-08T14:26:50+03:00GitHubnoreply@github.com2025-05-08T14:26:50+03:00context : remove logits_all flag (#13284)
178470a6991edf1b60e7afa8962f830320583f3babb0f0610212069929f92d428ae3b5596bfbe69c020bDiego Devesaslarengh@gmail.com2025-05-08T13:15:28+02:00GitHubnoreply@github.com2025-05-08T13:15:28+02:00ci : move release workflow to a separate file (#13362)
1785f0610212069929f92d428ae3b5596bfbe69c020b8733e0cf6eefc7c7752297cc22d0836706f4222cDiego Devesaslarengh@gmail.com2025-05-08T13:15:15+02:00GitHubnoreply@github.com2025-05-08T13:15:15+02:00llama : print size and type of overridden tensors (#13364)
17868733e0cf6eefc7c7752297cc22d0836706f4222c814f795e063c257f33b921eab4073484238a151aAlberto Cabrera Pérezalberto.cabrera@codeplay.com2025-05-08T10:08:01+01:00GitHubnoreply@github.com2025-05-08T10:08:01+01:00sycl: addressing non-contiguous src1 mul_mats (nc and batched) (#13343)
1787814f795e063c257f33b921eab4073484238a151ad879433824ac3c16b3b5f00075895d0ee9688e34Diego Devesaslarengh@gmail.com2025-05-07T16:36:33+02:00GitHubnoreply@github.com2025-05-07T16:36:33+02:00docker : disable arm64 and intel images (#13356)
1788d879433824ac3c16b3b5f00075895d0ee9688e3413b0a04597a4581cad4d9027a848f450c623801dGeorgi Gerganovggerganov@gmail.com2025-05-07T16:39:36+03:00Georgi Gerganovggerganov@gmail.com2025-05-07T17:28:36+03:00sync : ggml
178913b0a04597a4581cad4d9027a848f450c623801dbba9d945c14b93c5264b7956e00736601ca6f89aDaniel Beveniusdaniel.bevenius@gmail.com2025-05-05T13:09:35+02:00Georgi Gerganovggerganov@gmail.com2025-05-07T17:28:36+03:00whisper: remove MSVC warnings pragmas (whisper/3090)
1790bba9d945c14b93c5264b7956e00736601ca6f89abc4e1128f78be0fbb4e2fa630adb6a04b969ac68Jared Tweedjaredtwe@gmail.com2025-05-02T02:41:35-07:00Georgi Gerganovggerganov@gmail.com2025-05-07T17:28:36+03:00cmake : removed stdc++fs (whisper/3097)
1791bc4e1128f78be0fbb4e2fa630adb6a04b969ac6839e73ae0d69f882d7e29cecc6dd8f5052fca6731Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-07T12:49:27+02:00GitHubnoreply@github.com2025-05-07T12:49:27+02:00llama : deci : support ffn-free with attention (#13296)
179239e73ae0d69f882d7e29cecc6dd8f5052fca67311f73301b63668d61b5f3109489050a27dc3f65beYcros18012+ycros@users.noreply.github.com2025-05-07T18:23:28+10:00GitHubnoreply@github.com2025-05-07T11:23:28+03:00common : Add a warning when we can't match samplers from a string or char. (#13330)
17931f73301b63668d61b5f3109489050a27dc3f65be4773d7a02ffdb05ba9e673ff21ce95351836e33aR0CKSTARxiaodong.ye@mthreads.com2025-05-07T15:48:23+08:00GitHubnoreply@github.com2025-05-07T09:48:23+02:00cuda : remove nrows_x in mul_mat_q_process_tile (#13325)
17944773d7a02ffdb05ba9e673ff21ce95351836e33a6c7fd67b647a76846d1691cd181011dff4549d02Georgi Gerganovggerganov@gmail.com2025-05-07T10:28:02+03:00GitHubnoreply@github.com2025-05-07T10:28:02+03:00examples : remove infill (#13283)
17956c7fd67b647a76846d1691cd181011dff4549d02141a908a59bbc68ceae3bf090b850e33322a2ca9piDack104877312+piDack@users.noreply.github.com2025-05-07T15:23:11+08:00GitHubnoreply@github.com2025-05-07T09:23:11+02:00llama : support tie embedding for chatglm models (#13328)
1796141a908a59bbc68ceae3bf090b850e33322a2ca932916a49072f01c43e20df374af5f8a1f70d6963Johannes Gäßlerjohannesg@5d6.de2025-05-06T23:35:51+02:00GitHubnoreply@github.com2025-05-06T23:35:51+02:00CUDA: mix virt/real CUDA archs for GGML_NATIVE=OFF (#13135)
179732916a49072f01c43e20df374af5f8a1f70d6963ffc727203af1061fdeb49efef30f76171722e403Xuan-Son Nguyenson@huggingface.co2025-05-06T22:40:24+02:00GitHubnoreply@github.com2025-05-06T22:40:24+02:00clip : refactor graph builder (#13321)
1798ffc727203af1061fdeb49efef30f76171722e40391a86a6f354aa73a7aab7bc3d283be410fdc93a5DocShotgun126566557+DocShotgun@users.noreply.github.com2025-05-06T13:36:24-07:00GitHubnoreply@github.com2025-05-06T22:36:24+02:00sampling : make top_n_sigma no-op at <=0 or a single candidate (#13345)
179991a86a6f354aa73a7aab7bc3d283be410fdc93a5f4ed10b69cc38c54070a47f841827de5e8984cdfoobaboogaoobabooga4@gmail.com2025-05-06T15:24:15-03:00GitHubnoreply@github.com2025-05-06T20:24:15+02:00sampling : don't consider -infinity values in top_n_sigma (#13344)
1800f4ed10b69cc38c54070a47f841827de5e8984cdf1e333d5bba18e99bc328bb87ac1ee6a4e6260e0eDiego Devesaslarengh@gmail.com2025-05-06T20:15:31+02:00GitHubnoreply@github.com2025-05-06T20:15:31+02:00cmake : remove arm64 msvc presets (#13342)
18011e333d5bba18e99bc328bb87ac1ee6a4e6260e0e2f54e348ad2999c4e31b8777592247622b20420fAkarshan Biswasakarshan@menlo.ai2025-05-06T20:27:06+05:30GitHubnoreply@github.com2025-05-06T20:27:06+05:30SYCL: Disable reorder optimize by default and stop setting tensor extras when optimize is disabled (#13254)
18022f54e348ad2999c4e31b8777592247622b20420f2356fb1d53c86d838756211010bbabfafda7cb94Xuan-Son Nguyenson@huggingface.co2025-05-06T14:25:40+02:00GitHubnoreply@github.com2025-05-06T14:25:40+02:00llama : fix build_ffn without gate (#13336)
18032356fb1d53c86d838756211010bbabfafda7cb94764b85627b46f43d7ea801867cd1b6abef484574Johannes Gäßlerjohannesg@5d6.de2025-05-06T13:58:51+02:00GitHubnoreply@github.com2025-05-06T13:58:51+02:00CUDA: fix bad asserts for partial offload (#13337)
1804764b85627b46f43d7ea801867cd1b6abef48457415a28ec8c705b188ebe178170966d1dcc36fe151Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-06T11:12:06+02:00GitHubnoreply@github.com2025-05-06T11:12:06+02:00convert : qwen2/3moe : set yarn metadata if present (#13331)
180515a28ec8c705b188ebe178170966d1dcc36fe151a7366faa5bb2fff97b9fb43340d853709f52d8c9Johannes Gäßlerjohannesg@5d6.de2025-05-06T08:36:46+02:00GitHubnoreply@github.com2025-05-06T08:36:46+02:00CUDA: fix --split-mode row for MMQ (#13323)
1806a7366faa5bb2fff97b9fb43340d853709f52d8c9907036502070ba608bdb2aaebf802092d4cfba07compiladegit@compilade.net2025-05-05T22:27:31-04:00GitHubnoreply@github.com2025-05-05T22:27:31-04:00gguf-py : avoid requiring pyside6 for other scripts (#13036)
1807907036502070ba608bdb2aaebf802092d4cfba07233461f8121455f957a47e6a22a77b3bc88277b0Johannes Gäßlerjohannesg@5d6.de2025-05-05T22:32:13+02:00GitHubnoreply@github.com2025-05-05T22:32:13+02:00CUDA: fix logic for clearing padding with -ngl 0 (#13320)
1808233461f8121455f957a47e6a22a77b3bc88277b0b34c859146630dff136943abc9852ca173a7c9d6oobaboogaoobabooga4@gmail.com2025-05-05T17:12:19-03:00GitHubnoreply@github.com2025-05-05T22:12:19+02:00sampling : Integrate Top-nσ into main sampling chain (and add it to the server) (#13264)
1809b34c859146630dff136943abc9852ca173a7c9d69b61acf06041dcbaff6afa5f28940e93297f8520igardev49397134+igardev@users.noreply.github.com2025-05-05T17:03:31+03:00GitHubnoreply@github.com2025-05-05T16:03:31+02:00server : Webui - change setText command from parent window to also send the message. (#13309)
18109b61acf06041dcbaff6afa5f28940e93297f85205215b91e9377ce23e4ccc92ec3156bf5c7f892a3Xuan-Son Nguyenson@huggingface.co2025-05-05T16:02:55+02:00GitHubnoreply@github.com2025-05-05T16:02:55+02:00mtmd : rename llava directory to mtmd (#13311)
18115215b91e9377ce23e4ccc92ec3156bf5c7f892a3ae803bfc3d0fc2d0d3e1cce22ee103a30939e104Xuan-Son Nguyenson@huggingface.co2025-05-05T12:54:44+02:00GitHubnoreply@github.com2025-05-05T12:54:44+02:00clip : fix confused naming ffn_up and ffn_down (#13290)
1812ae803bfc3d0fc2d0d3e1cce22ee103a30939e10466645a5285d8c4c5f9a3b3f360d042baac2d820aSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-05T12:34:26+02:00GitHubnoreply@github.com2025-05-05T12:34:26+02:00convert : bailingmoe : set yarn metadata if present (#13312)
181366645a5285d8c4c5f9a3b3f360d042baac2d820a27aa2595321c4d9cc4086a8e67bdea204b8309b0Akarshan Biswasakarshan@menlo.ai2025-05-05T13:39:10+05:30GitHubnoreply@github.com2025-05-05T13:39:10+05:30SYCL: Disable mul_mat kernels for noncontiguous tensor b (#13308)
181427aa2595321c4d9cc4086a8e67bdea204b8309b09fdfcdaeddd1ef57c6d041b89cd8fb7048a0f028Xuan-Son Nguyenson@huggingface.co2025-05-04T23:43:42+02:00GitHubnoreply@github.com2025-05-04T23:43:42+02:00mtmd : add C public API (#13184)
18159fdfcdaeddd1ef57c6d041b89cd8fb7048a0f0286eb7d25c700e29d9272064db408855178d01741bDiego Devesaslarengh@gmail.com2025-05-04T21:25:43+02:00GitHubnoreply@github.com2025-05-04T21:25:43+02:00rpc : use backend registry, support dl backends (#13304)
18166eb7d25c700e29d9272064db408855178d01741b86bd60d3fe4a08b8c9d920e6defbc2412d803569Aaron Teoaaron.teo1@ibm.com2025-05-05T01:49:12+08:00GitHubnoreply@github.com2025-05-04T19:49:12+02:00ggml : activate s390x simd for Q3_K (#13301)
181786bd60d3fe4a08b8c9d920e6defbc2412d8035699f2da5871f4bbd205b8a3b952cdc76283218d595Diego Devesaslarengh@gmail.com2025-05-04T17:05:20+02:00GitHubnoreply@github.com2025-05-04T17:05:20+02:00llava/mtmd : fixes to fully support dl backends (#13303)
18189f2da5871f4bbd205b8a3b952cdc76283218d59593c4e23905987949b714b21ae918ff6bfb55fe36Diego Devesaslarengh@gmail.com2025-05-04T14:20:49+02:00GitHubnoreply@github.com2025-05-04T14:20:49+02:00llama : build windows releases with dl backends (#13220)
181993c4e23905987949b714b21ae918ff6bfb55fe368afbd968182909cf93fb15959fc867b6dd3adb53Johannes Gäßlerjohannesg@5d6.de2025-05-04T14:16:39+02:00GitHubnoreply@github.com2025-05-04T14:16:39+02:00CUDA: fix race condition in MMQ stream-k fixup (#13299)
18208afbd968182909cf93fb15959fc867b6dd3adb538ae5ebcf859b05a2ea3bbd930133a2fe4a89ed3cJohannes Gäßlerjohannesg@5d6.de2025-05-04T13:58:38+02:00GitHubnoreply@github.com2025-05-04T13:58:38+02:00CUDA: fix race condition in MMQ ids_dst (#13294)
18218ae5ebcf859b05a2ea3bbd930133a2fe4a89ed3c3e959f09764a2bb0e64af594eab83f7fb3e08eb2Jeff Bolzjbolz@nvidia.com2025-05-04T00:17:16-05:00GitHubnoreply@github.com2025-05-04T07:17:16+02:00vulkan: Additional type support for unary, binary, and copy (#13266)
18223e959f09764a2bb0e64af594eab83f7fb3e08eb236667c8edcded08063ed51c7d57e9e086bbfc903Johannes Gäßlerjohannesg@5d6.de2025-05-04T00:50:37+02:00GitHubnoreply@github.com2025-05-04T00:50:37+02:00imatrix: fix oob writes if src1 is not contiguous (#13286)
182336667c8edcded08063ed51c7d57e9e086bbfc9033bf785f3efa89ed28294fbf73054558a2b034bfbXuan-Son Nguyenson@huggingface.co2025-05-03T20:07:54+02:00GitHubnoreply@github.com2025-05-03T20:07:54+02:00clip : revert the change of BOI/EOI token for GLM-edge (⚠️ breaking change) (#13259)
18243bf785f3efa89ed28294fbf73054558a2b034bfb1d36b3670b285e69e58b9d687c770a2a0a192194ymcki84055651+ymcki@users.noreply.github.com2025-05-03T23:39:51+08:00GitHubnoreply@github.com2025-05-03T17:39:51+02:00llama : Llama-3_1-Nemotron-Ultra-253B-v1 support (#12843)
18251d36b3670b285e69e58b9d687c770a2a0a192194b34443923cad751483cc53af2e680d595daadce7Diego Devesaslarengh@gmail.com2025-05-02T20:27:13+02:00GitHubnoreply@github.com2025-05-02T20:27:13+02:00llama : move end-user examples to tools directory (#13249)
1826b34443923cad751483cc53af2e680d595daadce7a75cb30dc9e63488c3614e2d5a9fe2306eaf47cdGeorgi Gerganovggerganov@gmail.com2025-05-02T20:54:30+03:00GitHubnoreply@github.com2025-05-02T20:54:30+03:00sync : ggml (#13268)
1827a75cb30dc9e63488c3614e2d5a9fe2306eaf47cd3f3769ba76061a511f02f2a48da2ad2d93fce511Georgi Gerganovggerganov@gmail.com2025-05-02T20:54:13+03:00GitHubnoreply@github.com2025-05-02T20:54:13+03:00context : fix reorder logic (#13267)
18283f3769ba76061a511f02f2a48da2ad2d93fce5112f567611c0234bbca0a4009762acb47b56866095shalinib-ibmShalini.Salomi.Bodapati@ibm.com2025-05-02T22:23:12+05:30GitHubnoreply@github.com2025-05-02T19:53:12+03:00ggml : Enable MMA for BF16 in llamafile_sgemm (#13148)
18292f567611c0234bbca0a4009762acb47b568660957d2123484e6ba8fcd90fff8c01661b9dbbefdc9dJared Van Borteljared@nomic.ai2025-05-02T11:42:30-04:00GitHubnoreply@github.com2025-05-02T11:42:30-04:00llama-model : support Qwen2 embedding models and pooling_mode_lasttoken (#13245)
18307d2123484e6ba8fcd90fff8c01661b9dbbefdc9d074e42ab31de4c99aa7d9d2d239660f64b2380d6Jared Van Borteljared@nomic.ai2025-05-02T11:41:54-04:00GitHubnoreply@github.com2025-05-02T11:41:54-04:00convert : use correct context length for nomic-embed-text-v2 (#13216)
1831074e42ab31de4c99aa7d9d2d239660f64b2380d6c642bc014c105728ce45015813351dc5a37f60a2Xuan-Son Nguyenson@huggingface.co2025-05-02T17:17:15+02:00GitHubnoreply@github.com2025-05-02T17:17:15+02:00convert : converting mmproj for Qwen2/2.5VL from convert_hf_to_gguf (#13209)
1832c642bc014c105728ce45015813351dc5a37f60a2cb06a3c363f50cd35113984fe8fb164aea419077Georgi Gerganovggerganov@gmail.com2025-05-02T17:48:36+03:00GitHubnoreply@github.com2025-05-02T17:48:36+03:00kv-cache : separate recurrent vs non-recurrent impl (#12799)
1833cb06a3c363f50cd35113984fe8fb164aea419077626083faf73faa54440f934bb1741bf443be91b1Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-02T12:44:24+02:00GitHubnoreply@github.com2025-05-02T12:44:24+02:00llama : orion rope type is neox (#13261)
1834626083faf73faa54440f934bb1741bf443be91b12af6880178b4bc2c0eced726bab68b4bf333042bSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-05-02T12:40:56+02:00GitHubnoreply@github.com2025-05-02T12:40:56+02:00llama : plamo rope type is neox (#13260)
18352af6880178b4bc2c0eced726bab68b4bf333042be84773ab604fe0d935d03741df003716398dc57bpiDack104877312+piDack@users.noreply.github.com2025-05-02T17:06:09+08:00GitHubnoreply@github.com2025-05-02T11:06:09+02:00llama-chat : reset glmedge chat template (#13253)
1836e84773ab604fe0d935d03741df003716398dc57bfab647e8842c5f80da7e8f2c625dab6a0e19e5d4Shakil Ahmed44522075+ahmedshakill@users.noreply.github.com2025-05-02T14:20:27+06:00GitHubnoreply@github.com2025-05-02T10:20:27+02:00mtmd-cli : fix out_of_range when input image path is empty (#13244)
1837fab647e8842c5f80da7e8f2c625dab6a0e19e5d4dcf886007de4b8e5200f461a13233315f897fb9dGeorgi Gerganovggerganov@gmail.com2025-05-02T09:48:31+03:00GitHubnoreply@github.com2025-05-02T09:48:31+03:00server : add cache reuse card link to help (#13230)
1838dcf886007de4b8e5200f461a13233315f897fb9dd24d5928086471063fa9d9fd45aca710fd1336aeXuan-Son Nguyenson@huggingface.co2025-05-02T08:45:10+02:00GitHubnoreply@github.com2025-05-02T08:45:10+02:00convert : explicitly disable trust_remote_code for AutoConfig (#13246)
1839d24d5928086471063fa9d9fd45aca710fd1336ae8efbdadc616fa717c369059b9b388160958d886cbandoti141645996+bandoti@users.noreply.github.com2025-05-01T19:06:39-03:00GitHubnoreply@github.com2025-05-01T19:06:39-03:00ci: fix cross-compile sync issues (#12804)
18408efbdadc616fa717c369059b9b388160958d886cf057808ffadce213f54c1884ab5096e60140f358Justin Santa Barbarajustinsb@google.com2025-05-01T17:32:11-04:00GitHubnoreply@github.com2025-05-01T23:32:11+02:00rpc : avoid uninitialized memory in serialize_tensor (#13210)
1841f057808ffadce213f54c1884ab5096e60140f358d7a14c42a1883a34a6553cbfe30da1e1b84dfd6aJesse Grossjesse@kernel.org2025-05-01T13:46:10-07:00GitHubnoreply@github.com2025-05-01T22:46:10+02:00ggml: Don't assert fail when tensor data changes (#13222)
1842d7a14c42a1883a34a6553cbfe30da1e1b84dfd6ab6e4ff69b8abd509647b531bd5b4e86950204f66Diego Devesaslarengh@gmail.com2025-05-01T21:48:08+02:00GitHubnoreply@github.com2025-05-01T21:48:08+02:00build : fix build info on windows (#13239)
1843b6e4ff69b8abd509647b531bd5b4e86950204f66e0f572c8466e70d35cbd70ee536ad8fc83b2acacLoïc Carrèreloic.carrere@gmail.com2025-05-01T21:32:21+02:00GitHubnoreply@github.com2025-05-01T21:32:21+02:00clip : (minicpmv) Re-enable upscaling of images smaller than the CLIP image size (#13237)
1844e0f572c8466e70d35cbd70ee536ad8fc83b2acac79f26e9e125b21760aeb016f34bfd42a93f48351matteomatteo.serva@gmail.com2025-05-01T21:16:38+02:00GitHubnoreply@github.com2025-05-01T21:16:38+02:00llama-chat : update GLM4 chat template (#13238)
184579f26e9e125b21760aeb016f34bfd42a93f48351fc727bcdd5a311c7c69a76dbf87f4784e828c7b4Jeff Bolzjbolz@nvidia.com2025-05-01T13:49:39-05:00GitHubnoreply@github.com2025-05-01T20:49:39+02:00vulkan: Add bfloat16 support (#12554)
1846fc727bcdd5a311c7c69a76dbf87f4784e828c7b4b0ecbd434be024c06bf547491be444ed92e1123eJeff Bolzjbolz@nvidia.com2025-05-01T13:19:31-05:00GitHubnoreply@github.com2025-05-01T20:19:31+02:00vulkan: Handle src1 batch dimension in non-contiguous mat-vec-mul shader (#13191)
1847b0ecbd434be024c06bf547491be444ed92e1123eb1dd4d08e8fe40c9484422c0c5ec9bbd13b067a9Johannes Gäßlerjohannesg@5d6.de2025-05-01T20:18:56+02:00GitHubnoreply@github.com2025-05-01T20:18:56+02:00test: non-cont. b in test-backend-ops -o MUL_MAT (#13187)
1848b1dd4d08e8fe40c9484422c0c5ec9bbd13b067a999881f77d82efda80b21057d84f1cc2df2f1e0f6Georgi Gerganovggerganov@gmail.com2025-05-01T17:07:13+03:00Georgi Gerganovggerganov@gmail.com2025-05-01T20:15:34+03:00sync : ggml
184999881f77d82efda80b21057d84f1cc2df2f1e0f6b5769d92b4510c77691ad9e3f8b643c2ba202e53Daniel Beveniusdaniel.bevenius@gmail.com2025-05-01T10:05:24+02:00Georgi Gerganovggerganov@gmail.com2025-05-01T20:15:34+03:00whisper : add check that target name exists (whisper/3103)
1850b5769d92b4510c77691ad9e3f8b643c2ba202e538936784f7a1ec4f91637d04b77fdc90ec36ebac9Daniel Beveniusdaniel.bevenius@gmail.com2025-04-29T15:47:55+02:00Georgi Gerganovggerganov@gmail.com2025-05-01T20:15:34+03:00ggml : suppress Windows compiler warnings (whisper/3075)
18518936784f7a1ec4f91637d04b77fdc90ec36ebac913c9a3319b65469e883c49dd1c478abedc410157Xuan-Son Nguyenson@huggingface.co2025-05-01T17:05:42+02:00GitHubnoreply@github.com2025-05-01T17:05:42+02:00mtmd : add **vision** support for Mistral Small 3.1 (#13231)
185213c9a3319b65469e883c49dd1c478abedc410157a70183eb0079fdf6ebeaed12966338a039461b5dXuan-Son Nguyenson@huggingface.co2025-05-01T10:23:25+02:00GitHubnoreply@github.com2025-05-01T10:23:25+02:00arg : remove CURLINFO_EFFECTIVE_METHOD (#13228)
1853a70183eb0079fdf6ebeaed12966338a039461b5d8d33d740c308fe0049515668aac0e561269afe9eJared Van Borteljared@nomic.ai2025-05-01T03:09:41-04:00GitHubnoreply@github.com2025-05-01T10:09:41+03:00llama-model : fix the reported size class for nomic-embed-text-v2-moe (#13223)
18548d33d740c308fe0049515668aac0e561269afe9e4254bb49518e0f920f14c9aadd96eefdfd38b429Georgi Gerganovggerganov@gmail.com2025-05-01T09:59:02+03:00Georgi Gerganovggerganov@gmail.com2025-05-01T10:00:39+03:00sync : ggml
18554254bb49518e0f920f14c9aadd96eefdfd38b4299998540149a490200894acfe595e9a1546bab723Diego Devesaslarengh@gmail.com2025-04-30T15:20:40+02:00Georgi Gerganovggerganov@gmail.com2025-05-01T09:58:44+03:00ggml : fix ggml_gallocr_ptr type (ggml/1205)
18569998540149a490200894acfe595e9a1546bab723e1e8e0991ffd9e99a445c6812bb519d5bac9f4b5Georgi Gerganovggerganov@gmail.com2025-04-24T18:59:06+03:00Georgi Gerganovggerganov@gmail.com2025-05-01T09:58:44+03:00cuda : fix unused variable compile warning (whisper/0)
1857e1e8e0991ffd9e99a445c6812bb519d5bac9f4b56f67cf1f480926391ad75ff746e0a021214bf70cJohannes Gäßlerjohannesg@5d6.de2025-04-30T23:12:59+02:00GitHubnoreply@github.com2025-04-30T23:12:59+02:00CUDA: batched+noncont MMQ, refactor bs>1 MoE code (#13199)
18586f67cf1f480926391ad75ff746e0a021214bf70c16a457facd996915652f6274384c87602b27d21aXuan-Son Nguyenson@huggingface.co2025-04-30T22:29:15+02:00GitHubnoreply@github.com2025-04-30T21:29:15+01:00arg : -hf do not fail if url mismatch (#13219)
185916a457facd996915652f6274384c87602b27d21a3e168bede4d27b35656ab8026015b87659ecbec2ddh0dylanhalladay02@icloud.com2025-04-30T15:28:43-05:00GitHubnoreply@github.com2025-04-30T21:28:43+01:00fix typo: `n_ctx_pre_seq` -> `n_ctx_per_seq` (#13221)
18603e168bede4d27b35656ab8026015b87659ecbec2ceda28ef8e310a8dee60bf275077a3eedae8e36cXuan-Son Nguyenson@huggingface.co2025-04-30T16:56:24+02:00GitHubnoreply@github.com2025-04-30T16:56:24+02:00convert : improve model arch handling (#13122)
1861ceda28ef8e310a8dee60bf275077a3eedae8e36c3b127c738535d95e06abd0d43da147bc13516ad0Tatsuya Tanakatanakasan2525@gmail.com2025-04-30T22:25:20+09:00GitHubnoreply@github.com2025-04-30T15:25:20+02:00llava : remove duplicate include (#13207)
18623b127c738535d95e06abd0d43da147bc13516ad0e5007a5edf2692ef7151a81a61ce2716b83374e5Olivier Chafikochafik@users.noreply.github.com2025-04-30T13:52:35+01:00GitHubnoreply@github.com2025-04-30T14:52:35+02:00common : add -jf / --json-schema-file flag (#12011)
1863e5007a5edf2692ef7151a81a61ce2716b83374e5416313773b53585fddcafbcb914cbbfbaeb94b1fJeff Bolzjbolz@nvidia.com2025-04-30T07:38:37-05:00GitHubnoreply@github.com2025-04-30T14:38:37+02:00vulkan: use uint array index to avoid glslang bug (#13193)
1864416313773b53585fddcafbcb914cbbfbaeb94b1f07c2e2f76cce9a61c110b6995fbb90ccea2c3aaashalinib-ibmShalini.Salomi.Bodapati@ibm.com2025-04-30T16:47:08+05:30GitHubnoreply@github.com2025-04-30T13:17:08+02:00ggml : fix ppc64le build (#13176)
186507c2e2f76cce9a61c110b6995fbb90ccea2c3aaa44cd8d91ff2c9e4a0f2e3151f8d6f04c928e2571Xuan-Son Nguyenson@huggingface.co2025-04-30T13:06:15+02:00GitHubnoreply@github.com2025-04-30T13:06:15+02:00convert : correct typo image_mean --> image_std (#13208)
186644cd8d91ff2c9e4a0f2e3151f8d6f04c928e25715933e6fdc9c051eea6c83b5a7608de12f9f15670Aaron Teo57927438+taronaeo@users.noreply.github.com2025-04-30T17:47:35+08:00GitHubnoreply@github.com2025-04-30T10:47:35+01:00feat(ggml-cpu): enable z17 compile (#13182)
18675933e6fdc9c051eea6c83b5a7608de12f9f15670da84c04d8fa43ff92b172feb8130c74d062f956aXuan-Son Nguyenson@huggingface.co2025-04-30T10:46:32+02:00GitHubnoreply@github.com2025-04-30T10:46:32+02:00arg : allow using -hf offline (#13202)
1868da84c04d8fa43ff92b172feb8130c74d062f956aa0f7016d170ca4bfe24d9a9f26c024d034af69f2Xuan-Son Nguyenson@huggingface.co2025-04-30T10:44:07+02:00GitHubnoreply@github.com2025-04-30T10:44:07+02:00docker : do not build tests (#13204)
1869a0f7016d170ca4bfe24d9a9f26c024d034af69f219e899ce21a7c9ffcf8bb2b22269a75f6e078f8fxiaofeihbuxiaofei@gmail.com2025-04-30T14:29:22+08:00GitHubnoreply@github.com2025-04-30T09:29:22+03:00rpc : fix cache directory initialization (#13188)
187019e899ce21a7c9ffcf8bb2b22269a75f6e078f8fe2e1ddb93a01ce282e304431b37e60b3cddb6114Johannes Gäßlerjohannesg@5d6.de2025-04-29T23:32:04+02:00GitHubnoreply@github.com2025-04-29T23:32:04+02:00scripts: n_depth for compare-llama-bench [no ci] (#13201)
1871e2e1ddb93a01ce282e304431b37e60b3cddb6114d9d398f84f96d16c308d4976f5e90222ecc2a492matteomatteo.serva@gmail.com2025-04-29T20:33:10+02:00GitHubnoreply@github.com2025-04-29T20:33:10+02:00server : Prefilling assistant message in openai compatible API (#13174)
1872d9d398f84f96d16c308d4976f5e90222ecc2a4925a6398011704c31178d7b774be67856ba57647c8Georgi Gerganovggerganov@gmail.com2025-04-29T20:22:57+03:00GitHubnoreply@github.com2025-04-29T20:22:57+03:00sampling : when top-k <= 0 -> noop (#13173)
18735a6398011704c31178d7b774be67856ba57647c8cdf76586b23c67abd3ca064ee2c084c57ae240bdAlberto Cabrera Pérezalberto.cabrera@codeplay.com2025-04-29T16:24:36+01:00GitHubnoreply@github.com2025-04-29T17:24:36+02:00llama-bench: fixed size of fields to correctly map to values (#13183)
1874cdf76586b23c67abd3ca064ee2c084c57ae240bd7d3af70b089bb349b5d17eb01839224c99ec1952Johannes Gäßlerjohannesg@5d6.de2025-04-29T16:00:27+02:00GitHubnoreply@github.com2025-04-29T16:00:27+02:00CUDA: fix non-cont. inputs for batched mat mul (#13155)
18757d3af70b089bb349b5d17eb01839224c99ec195200e3e5a194e88e604e7c91391b9e90332888fd72Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-04-29T13:25:53+02:00GitHubnoreply@github.com2025-04-29T13:25:53+02:00llama : llm_type order by size (#13177)
187600e3e5a194e88e604e7c91391b9e90332888fd72e98b3692be4cd8fbbd9a56fbacc2f2bf0bf26a68Xuan-Son Nguyenson@huggingface.co2025-04-29T11:47:04+02:00GitHubnoreply@github.com2025-04-29T11:47:04+02:00mtmd : add qwen2vl and qwen2.5vl (#13141)
1877e98b3692be4cd8fbbd9a56fbacc2f2bf0bf26a68b6ce7430b7eb51f032152316880204e0a9c0470eSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-04-29T11:00:31+02:00GitHubnoreply@github.com2025-04-29T11:00:31+02:00llama : set qwen3 model type sizes (#13175)
1878b6ce7430b7eb51f032152316880204e0a9c0470e5f5e39e1ba5dbea814e41f2a15e035d749a520bcXuan-Son Nguyenson@huggingface.co2025-04-29T08:45:49+02:00GitHubnoreply@github.com2025-04-29T09:45:49+03:00llama-graph : fix text position for mrope (#13159)
18795f5e39e1ba5dbea814e41f2a15e035d749a520bceaea3253244dc4bbe07f6cd81325847ccc6cf93eATmanyoso@users.noreply.github.com2025-04-28T15:52:15-04:00GitHubnoreply@github.com2025-04-28T22:52:15+03:00model : Nomic Embed Text V2 with Mixture-of-Experts (MoE) architecture (#12466)
1880eaea3253244dc4bbe07f6cd81325847ccc6cf93e43ddab6eeeaab5a04fe5a364af0bafb0e4d35065Xuan-Son Nguyenson@huggingface.co2025-04-28T21:23:19+02:00GitHubnoreply@github.com2025-04-28T21:23:19+02:00clip : fix model size display (#13153)
188143ddab6eeeaab5a04fe5a364af0bafb0e4d350651831f538f720d1d99fba146f24f0a8e970838cc4Ville Vesilehtoville@vesilehto.fi2025-04-28T21:00:20+03:00GitHubnoreply@github.com2025-04-28T21:00:20+03:00fix(rpc): Improve input validation and error handling (#13069)
18821831f538f720d1d99fba146f24f0a8e970838cc44e87962e34a4b257ec374c4baf6b1568554b81a9Vishal Agarwalvishalagarwal.jss@gmail.com2025-04-28T20:20:39+05:30GitHubnoreply@github.com2025-04-28T16:50:39+02:00llama-bench: add `-d` depth arg (#13096)
18834e87962e34a4b257ec374c4baf6b1568554b81a9fb0471d1753824e75474c24f82fbdd54c94dcedaXuan-Son Nguyenson@huggingface.co2025-04-28T16:12:56+02:00GitHubnoreply@github.com2025-04-28T16:12:56+02:00mtmd : fix glm-edge redundant token count (#13139)
1884fb0471d1753824e75474c24f82fbdd54c94dcedad2b2031e5f11b826dcc718138642f147a2009665pockers21134406831+pockers21@users.noreply.github.com2025-04-28T06:45:40-07:00GitHubnoreply@github.com2025-04-28T16:45:40+03:00context : do not clear output buffer on reserve (#13152)
1885d2b2031e5f11b826dcc718138642f147a20096655fa9e63be82225fb3249c76f39ddda3e5bdec0a3Xuan-Son Nguyenson@huggingface.co2025-04-28T14:20:56+02:00GitHubnoreply@github.com2025-04-28T14:20:56+02:00llama : (mrope) allow using normal 1D position for text token (#13138)
18865fa9e63be82225fb3249c76f39ddda3e5bdec0a3a4c340f974f9b7ac0c1aae897aabaa54549a97e5Xuan-Son Nguyenson@huggingface.co2025-04-28T12:18:59+02:00GitHubnoreply@github.com2025-04-28T12:18:59+02:00clip : refactor set input for cgraph + fix qwen2.5vl input (#13136)
1887a4c340f974f9b7ac0c1aae897aabaa54549a97e5d0a417f3c7a5a22ef05b3b76d91dbe1d3362bf0cAkarshan Biswasakarshan@menlo.ai2025-04-28T15:03:25+05:30GitHubnoreply@github.com2025-04-28T11:33:25+02:00SYCL: Add all missing unary kernels (#13074)
1888d0a417f3c7a5a22ef05b3b76d91dbe1d3362bf0c43f2b07193cbcccd266734320ea9b948f5a01926Georgi Gerganovggerganov@gmail.com2025-04-28T12:10:18+03:00GitHubnoreply@github.com2025-04-28T12:10:18+03:00readme : update hot topics (#13150)
188943f2b07193cbcccd266734320ea9b948f5a01926e5d6c2554e7597665e26991a93fa2f3d16c79ad5Georgi Gerganovggerganov@gmail.com2025-04-28T11:57:19+03:00GitHubnoreply@github.com2025-04-28T11:57:19+03:00common : fix noreturn compile warning (#13151)
1890e5d6c2554e7597665e26991a93fa2f3d16c79ad5f0dd6a1926cdb2f4183a937deee40db26ef8f1daXuan-Son Nguyenson@huggingface.co2025-04-28T10:11:58+02:00GitHubnoreply@github.com2025-04-28T10:11:58+02:00llama-chat : fix typo GML --> GLM (#13143)
1891f0dd6a1926cdb2f4183a937deee40db26ef8f1da69699be48a6b94570773532850667f1591dc5bbeR0CKSTARxiaodong.ye@mthreads.com2025-04-28T15:33:28+08:00GitHubnoreply@github.com2025-04-28T09:33:28+02:00musa: fix typo in cc control (#13144)
189269699be48a6b94570773532850667f1591dc5bbe85f36e5e7173eef7c671c778db44c034e1d0ab19Johannes Gäßlerjohannesg@5d6.de2025-04-28T09:29:26+02:00GitHubnoreply@github.com2025-04-28T09:29:26+02:00CUDA: fix q_nope_absorbed prec for DS 2 Lite f16 (#13137)
189385f36e5e7173eef7c671c778db44c034e1d0ab19c0a97b762e5ec767dc414f0dc4979befd4c09a52Xuan-Son Nguyenson@huggingface.co2025-04-28T07:16:59+02:00GitHubnoreply@github.com2025-04-28T08:16:59+03:00arg : fix unused variable (#13142)
1894c0a97b762e5ec767dc414f0dc4979befd4c09a52ced44be34290fab450f8344efa047d8a08e723b44onen11580688+4onen@users.noreply.github.com2025-04-27T14:48:26-07:00GitHubnoreply@github.com2025-04-27T23:48:26+02:00llama-bench : Add `--override-tensors` arg (#12922)
1895ced44be34290fab450f8344efa047d8a08e723b4e291450b7602d7a36239e4ceeece37625f838373matteomatteo.serva@gmail.com2025-04-27T21:57:32+02:00GitHubnoreply@github.com2025-04-27T21:57:32+02:00llama-chat : fix wrong template in GLM4-0414 (#13140)
1896e291450b7602d7a36239e4ceeece37625f83837359e991c23cc44cb5fb657e7b9358cac21fb79828R0CKSTARxiaodong.ye@mthreads.com2025-04-27T19:22:49+08:00GitHubnoreply@github.com2025-04-27T13:22:49+02:00musa: fix build warning (#13129)
189759e991c23cc44cb5fb657e7b9358cac21fb79828ca2bb89eac2097ab4620448737e58af8452e444bLostRuins Concedo39025047+LostRuins@users.noreply.github.com2025-04-27T18:43:37+08:00GitHubnoreply@github.com2025-04-27T12:43:37+02:00Fixes Qwen2.5VL segfault during inference with https://github.com/ggml-org/llama.cpp/pull/12402 as has_qwen2vl_merger migration was incomplete (#13133)
1898ca2bb89eac2097ab4620448737e58af8452e444b2d451c80590b9ac250322769ac13d3b4870dbcf7HimariOdsfhe49854@gmail.com2025-04-27T16:10:34+08:00GitHubnoreply@github.com2025-04-27T10:10:34+02:00clip : Add Qwen2.5VL support (#12402)
18992d451c80590b9ac250322769ac13d3b4870dbcf74753791e70acd4d4e02f2098f14a03df26c992bdXuan-Son Nguyenson@huggingface.co2025-04-26T22:58:12+02:00GitHubnoreply@github.com2025-04-26T22:58:12+02:00common : add common_remote_get_content (#13123)
19004753791e70acd4d4e02f2098f14a03df26c992bd77d5e9a76a7b4a8a7c5bf9cf6ebef91860123cbaXuan-Son Nguyenson@huggingface.co2025-04-26T22:39:47+02:00GitHubnoreply@github.com2025-04-26T22:39:47+02:00clip : improve projector naming (#13118)
190177d5e9a76a7b4a8a7c5bf9cf6ebef91860123cbad5fe4e81bd447124836ecfb47d794f8768665b9fSXXsxx1136965276@gmail.com2025-04-26T22:05:31+08:00GitHubnoreply@github.com2025-04-26T16:05:31+02:00ggml: move fp16/bf16 conversion optimizations to CPU backend + export conversion APIs (#13107)
1902d5fe4e81bd447124836ecfb47d794f8768665b9f295354ea6848a77bdee204ee1c971d9b92ffcca9frobrick+github@frob.com.au2025-04-26T10:10:20+02:00GitHubnoreply@github.com2025-04-26T10:10:20+02:00grammar : handle maxItems == 0 in JSON schema (#13117)
1903295354ea6848a77bdee204ee1c971d9b92ffcca9558a764713468f26f5a163d25a22100c9a04a48fDiego Devesaslarengh@gmail.com2025-04-25T19:40:11+02:00GitHubnoreply@github.com2025-04-25T19:40:11+02:00llama : fix K-shift with quantized K and BLAS backend (#13113)
1904558a764713468f26f5a163d25a22100c9a04a48fedb18b6e8f5ea6509ad43057f8bb98fc557dbc4eCity125218114+city96@users.noreply.github.com2025-04-25T14:38:34+02:00GitHubnoreply@github.com2025-04-25T14:38:34+02:00Force FP32 compute in GLM4 FFN Down (#13101)
1905edb18b6e8f5ea6509ad43057f8bb98fc557dbc4e514c45608f93f66106a712dee1abe062099ce790Xuan-Son Nguyenson@huggingface.co2025-04-25T14:31:42+02:00GitHubnoreply@github.com2025-04-25T14:31:42+02:00clip : fix pixtral on some GPU backends (#13097)
1906514c45608f93f66106a712dee1abe062099ce790553a5c3a9fdf771be2101bc3529937963f817457Neo Zhang Jianyujianyu.zhang@intel.com2025-04-25T17:37:51+08:00GitHubnoreply@github.com2025-04-25T17:37:51+08:00change the reorder tensor from init to execute OP (#13003)
1907553a5c3a9fdf771be2101bc3529937963f81745713be08daf992c89d5169518229b3740041c0f419Radoslav Gerganovrgerganov@gmail.com2025-04-25T10:08:08+03:00GitHubnoreply@github.com2025-04-25T10:08:08+03:00rpc : do not wait for response when sending RPC_CMD_SET_TENSOR (#12943)
190813be08daf992c89d5169518229b3740041c0f419226251ed56b85190e18a1cca963c45b888f4953cXuan-Son Nguyenson@huggingface.co2025-04-24T22:17:04+02:00GitHubnoreply@github.com2025-04-24T22:17:04+02:00clip : remove boi/eoi embeddings for GLM-edge model (#13081)
1909226251ed56b85190e18a1cca963c45b888f4953c87616f0680947800ecba3e9f6bc6e101943bf8e6Georgi Gerganovggerganov@gmail.com2025-04-24T22:29:22+03:00GitHubnoreply@github.com2025-04-24T22:29:22+03:00embeddings : fix batch sizes (#13076)
191087616f0680947800ecba3e9f6bc6e101943bf8e663b4911494afe04778c61b9c19019341d71c99fcGeorgi Gerganovggerganov@gmail.com2025-04-24T17:22:27+03:00Georgi Gerganovggerganov@gmail.com2025-04-24T17:32:47+03:00ggml : fix trailing whitespaces (#0)
191163b4911494afe04778c61b9c19019341d71c99fcc6e8cc28c15166dba15629dba6a7366d4d5955caGeorgi Gerganovggerganov@gmail.com2025-04-24T16:47:43+03:00Georgi Gerganovggerganov@gmail.com2025-04-24T17:32:47+03:00sync : ggml
1912c6e8cc28c15166dba15629dba6a7366d4d5955cab10d8bfdb1dac40cce34e8860ca5ec7d950c3a44Aclyaclysia@gmail.com2025-04-17T14:16:45+02:00Georgi Gerganovggerganov@gmail.com2025-04-24T17:32:47+03:00ggml : Depthwise 2D convolution (ggml/1152)
1913b10d8bfdb1dac40cce34e8860ca5ec7d950c3a4413b4548877326fdabee3e831b8cfd65d9844383cJohannes Gäßlerjohannesg@5d6.de2025-04-24T15:57:10+02:00GitHubnoreply@github.com2025-04-24T15:57:10+02:00CUDA: use switch statements in constexpr functions (#13095)
191413b4548877326fdabee3e831b8cfd65d9844383c572b3141d343d7f947bf53b57513016e90db5680Georgi Gerganovggerganov@gmail.com2025-04-24T16:00:10+03:00GitHubnoreply@github.com2025-04-24T16:00:10+03:00cmake : do not include ./src as public for libllama (#13062)
1915572b3141d343d7f947bf53b57513016e90db56807c727fbe39150fbe8381f4fa43fed08719ebebe6Georgi Gerganovggerganov@gmail.com2025-04-24T15:44:05+03:00GitHubnoreply@github.com2025-04-24T15:44:05+03:00clang-tidy : disable warning about missing math parenthesis (#13091)
19167c727fbe39150fbe8381f4fa43fed08719ebebe680982e815e67bae2442237f4e11466f44c9a2988Xuan-Son Nguyenson@huggingface.co2025-04-24T14:04:14+02:00GitHubnoreply@github.com2025-04-24T14:04:14+02:00arg : add --no-mmproj-offload (#13093)
191780982e815e67bae2442237f4e11466f44c9a29887604a7d6b80e78eef8f275fc700d0f64820d672fXuan-Son Nguyenson@huggingface.co2025-04-24T12:14:13+02:00GitHubnoreply@github.com2025-04-24T12:14:13+02:00arg : clean up handling --mmproj with -hf (#13082)
19187604a7d6b80e78eef8f275fc700d0f64820d672fb3b6d862cfdf190e1b9ad961639a25f5ebc0c7e3Georgi Gerganovggerganov@gmail.com2025-04-24T10:38:30+03:00GitHubnoreply@github.com2025-04-24T10:38:30+03:00metal : fix floating-point range of attention scores in FA kernels (#13090)
1919b3b6d862cfdf190e1b9ad961639a25f5ebc0c7e356304069599f4dd9749d94b9bca2c2c65bb27c02Eve139727413+netrunnereve@users.noreply.github.com2025-04-24T07:18:33ZGitHubnoreply@github.com2025-04-24T09:18:33+02:00vulkan: matmul gcn tuning (#13016)
192056304069599f4dd9749d94b9bca2c2c65bb27c02ecda2ec4b347031a9b8a89ee2efc664ce63f599cpl752pl752@mail.ru2025-04-24T02:32:35+05:00GitHubnoreply@github.com2025-04-23T23:32:35+02:00llama-mtmd-cli: Sigint rework in mtmd vision example (#13080)
1921ecda2ec4b347031a9b8a89ee2efc664ce63f599ceb1776b15a32d832f1266deeeab75b9d255c5849Xuan-Son Nguyenson@huggingface.co2025-04-23T20:21:59+02:00GitHubnoreply@github.com2025-04-23T20:21:59+02:00mtmd : Support Pixtral 12B (#13065)
1922eb1776b15a32d832f1266deeeab75b9d255c58492cca6c01e46d2fc1124d15730273ed2acdad1016piDack104877312+piDack@users.noreply.github.com2025-04-23T22:59:14+08:00GitHubnoreply@github.com2025-04-23T16:59:14+02:00convert : Append mult-eos,half-rope,bos to GLM4-0414 and Z (#13021)
19232cca6c01e46d2fc1124d15730273ed2acdad1016658987cfc9d752dca7758987390d5fb1a7a0a54aRadoslav Gerganovrgerganov@gmail.com2025-04-23T10:32:49+03:00GitHubnoreply@github.com2025-04-23T10:32:49+03:00rpc : add command line option for number of threads for the CPU backend (#13060)
1924658987cfc9d752dca7758987390d5fb1a7a0a54adc39a5e7a84815a90fa0c515ed8927870cf858c9Johannes Gäßlerjohannesg@5d6.de2025-04-22T21:27:40+02:00GitHubnoreply@github.com2025-04-22T21:27:40+02:00CUDA: noncont MMVQ + batched bs1 MUL_MAT_ID (#13014)
1925dc39a5e7a84815a90fa0c515ed8927870cf858c9ab47dec3d37aa1927c2ec590e166b76141374ed3Xuan-Son Nguyenson@huggingface.co2025-04-22T16:24:54+02:00GitHubnoreply@github.com2025-04-22T16:24:54+02:00mtmd : support SmolVLM (version 1 and 2) (#13050)
1926ab47dec3d37aa1927c2ec590e166b76141374ed37b53389c24a507564be39e1ea82746a39749059bGeorgi Gerganovggerganov@gmail.com2025-04-22T16:16:10+03:00GitHubnoreply@github.com2025-04-22T16:16:10+03:00security : add note about RPC and server functionality (#13061)
19277b53389c24a507564be39e1ea82746a39749059b243453533e029334181dda50d911d5fc5a2b2486Georgi Gerganovggerganov@gmail.com2025-04-22T16:15:51+03:00GitHubnoreply@github.com2025-04-22T16:15:51+03:00metal : add memory pool for temp allocs (#12850)
1928243453533e029334181dda50d911d5fc5a2b24861d735c0b4fa0551c51c2f4ac888dd9a01f447985Xuan-Son Nguyenson@huggingface.co2025-04-22T10:37:00+02:00GitHubnoreply@github.com2025-04-22T10:37:00+02:00llava : update documentations (#13055)
19291d735c0b4fa0551c51c2f4ac888dd9a01f4479855368ddda7a262d195b54687a31009dcc1f8b1602Diego Devesaslarengh@gmail.com2025-04-21T18:13:51+02:00GitHubnoreply@github.com2025-04-21T18:13:51+02:00ggml : add SSE 4.2 and x64 base variant for CPUs without AVX (#12871)
19305368ddda7a262d195b54687a31009dcc1f8b160284a9bf2fc2875205f0806fbbfbb66dc67204094cAkarshan Biswasakarshan@menlo.ai2025-04-21T19:13:30+05:30GitHubnoreply@github.com2025-04-21T19:13:30+05:30SYCL: Add non-contiguous support in ROPE (#12993)
193184a9bf2fc2875205f0806fbbfbb66dc67204094c2016f07bd106c73699ecbaace80f55db5ed95dacXuan-Son Nguyenson@huggingface.co2025-04-21T15:32:58+02:00GitHubnoreply@github.com2025-04-21T15:32:58+02:00mtmd : merge llava, gemma3 and minicpmv CLI into single `llama-mtmd-cli` (#13012)
19322016f07bd106c73699ecbaace80f55db5ed95dac6602304814e679cc8c162bb760a034aceb4f8965Xuan-Son Nguyenson@huggingface.co2025-04-20T23:29:36+02:00GitHubnoreply@github.com2025-04-20T23:29:36+02:00convert : experimental support for `--mmproj` flag (#13023)
19336602304814e679cc8c162bb760a034aceb4f896566168204be9559dc841f06f0025f3da01d9a8546Jeffrey Morganjmorganca@gmail.com2025-04-20T03:15:41-07:00GitHubnoreply@github.com2025-04-20T12:15:41+02:00llava: fix errors in clip.h on certain compilers (#13030)
193466168204be9559dc841f06f0025f3da01d9a85464ba9d711ba0a5bf0be00936d3258d4a5e4164d4fJeff Bolzjbolz@nvidia.com2025-04-20T03:50:02-05:00GitHubnoreply@github.com2025-04-20T10:50:02+02:00vulkan: support noncontiguous rms_norm (#13031)
19354ba9d711ba0a5bf0be00936d3258d4a5e4164d4f00137157fca3d17b90380762b4d7cc158d385bd3Jeffrey Morganjmorganca@gmail.com2025-04-19T22:28:40-07:00GitHubnoreply@github.com2025-04-20T08:28:40+03:00metal: add neg operator (#13029)
193600137157fca3d17b90380762b4d7cc158d385bd3fb28f4f80efb5a2990a6a240433b02e259b0dbb1bandoti141645996+bandoti@users.noreply.github.com2025-04-19T13:05:03-03:00GitHubnoreply@github.com2025-04-19T18:05:03+02:00Disable CI cross-compile builds (#13022)
1937fb28f4f80efb5a2990a6a240433b02e259b0dbb137b9f0d29d7301dd0ec5dfae8f357ccee96325d7Sigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-04-19T16:26:38+02:00GitHubnoreply@github.com2025-04-19T16:26:38+02:00gguf-py : fix upload python package workflow (#13020)
193837b9f0d29d7301dd0ec5dfae8f357ccee96325d76408210082cc0a61b992b487be7e2ff2efbb9e36Xuan-Son Nguyenson@huggingface.co2025-04-19T09:15:45+02:00GitHubnoreply@github.com2025-04-19T09:15:45+02:00clip : refactor, add `image_manipulation` and `llava_uhd` classes (#13011)
19396408210082cc0a61b992b487be7e2ff2efbb9e36aff9d107b066c9d464f7ab1324280acfdcebf569Daniel Tangdanielzgtg.opensource@gmail.com2025-04-18T16:02:55-04:00GitHubnoreply@github.com2025-04-18T22:02:55+02:00main : Fix Ctrl+D/newline handling (#12951)
1940aff9d107b066c9d464f7ab1324280acfdcebf56935370ba94593c3abc9550328377d461fc4f822b7Chris Thompsonchristopherthompson81@gmail.com2025-04-18T12:30:41-06:00GitHubnoreply@github.com2025-04-18T20:30:41+02:00gguf-py : GGUF Editor GUI - Python + Qt6 (#12930)
194135370ba94593c3abc9550328377d461fc4f822b78d6600576318dfc6b091fca744b0fd36a5e5255fXuan-Son Nguyenson@huggingface.co2025-04-18T19:58:12+02:00GitHubnoreply@github.com2025-04-18T19:58:12+02:00server : use std::move whenever possible (#12936)
19428d6600576318dfc6b091fca744b0fd36a5e5255fb9154ecff93ff54dc554411eb844a2a654be49f2Akarshan Biswasakarshan@menlo.ai2025-04-18T19:27:56+05:30GitHubnoreply@github.com2025-04-18T15:57:56+02:00SYCL: Refactor and enable FP16 in binary broadcast OPs (#12975)
1943b9154ecff93ff54dc554411eb844a2a654be49f22db9ba1464f3de0aceb2b5289963e69fc369cb66Xuan-Son Nguyenson@huggingface.co2025-04-18T10:04:51+02:00GitHubnoreply@github.com2025-04-18T10:04:51+02:00mtmd : add methods to access `mtmd_image_tokens` (#12906)
19442db9ba1464f3de0aceb2b5289963e69fc369cb662f74c354c0f752ed9aabf7d3a350e6edebd7e744Radoslav Gerganovrgerganov@gmail.com2025-04-18T10:13:42+03:00GitHubnoreply@github.com2025-04-18T10:13:42+03:00rpc : add RPC_CMD_HELLO (#12955)
19452f74c354c0f752ed9aabf7d3a350e6edebd7e744207c22ec2d6d793fc70830138617d1e016c5151cGeorgi Gerganovggerganov@gmail.com2025-04-17T18:16:36+03:00GitHubnoreply@github.com2025-04-17T18:16:36+03:00graph : make FA compatible with MLA + add initial Metal kernels (#12953)
1946207c22ec2d6d793fc70830138617d1e016c5151c7a395f67a7a02bb361d944b816d6e933889e28e1Alan Grayagray3@users.noreply.github.com2025-04-17T14:19:42+01:00GitHubnoreply@github.com2025-04-17T15:19:42+02:00ggml: Re-enable CUDA graphs in presence of CONT and DUP nodes (#12970)
19477a395f67a7a02bb361d944b816d6e933889e28e1971f245b3b5f3f55991bb779cb541b00f82eea1dhipuddinghuafengchun@gmail.com2025-04-17T20:34:16+08:00GitHubnoreply@github.com2025-04-17T20:34:16+08:00CANN: Add support for async operator submission (#12864)
1948971f245b3b5f3f55991bb779cb541b00f82eea1d12b17501e6015ffe568ac54fdf08e6580833bf1bMikko Juolamikjuo@gmail.com2025-04-17T01:37:05-07:00GitHubnoreply@github.com2025-04-17T11:37:05+03:00llama : recognize IBM Granite 3.3 FIM tokens (#12988)
194912b17501e6015ffe568ac54fdf08e6580833bf1b015022bb53387baa8b23817ac03743705c7d472bkimminsu80271594+kimminsu38oo@users.noreply.github.com2025-04-17T06:25:57+09:00GitHubnoreply@github.com2025-04-16T14:25:57-07:00opencl: fix incorrect local_size index in profiling log (#12868)
1950015022bb53387baa8b23817ac03743705c7d472bb43d89e311c5e7fbf62e5ec3c0401eb536677267Jeff Bolzjbolz@nvidia.com2025-04-16T13:37:25-05:00GitHubnoreply@github.com2025-04-16T20:37:25+02:00vulkan: enable coopmat2 FA gqa and split_k optimizations more often (#12931)
1951b43d89e311c5e7fbf62e5ec3c0401eb53667726780f19b41869728eeb6a26569957b92a773a2b2c6Chenguang Li757486878@qq.com2025-04-16T16:21:05+08:00GitHubnoreply@github.com2025-04-16T16:21:05+08:00CANN: Add 310P operator support check (#12962)
195280f19b41869728eeb6a26569957b92a773a2b2c6f8f820cc4dc37032d5375972ba904ce53043445dlhezquic_lih@quicinc.com2025-04-15T12:26:00-07:00GitHubnoreply@github.com2025-04-15T12:26:00-07:00opencl: split `ggml-opencl.cl` into multiple files and cleanup (#12886)
1953f8f820cc4dc37032d5375972ba904ce53043445d54a72720432810c89c2693f34908b60b88da1e46Georgi Gerganovggerganov@gmail.com2025-04-15T14:45:05+03:00GitHubnoreply@github.com2025-04-15T14:45:05+03:00metal : add FA-vec kernels for head size 96 (#12952)
195454a72720432810c89c2693f34908b60b88da1e4684778e97703740d8ac5fb64e14d83b80eafa0f3chipuddinghuafengchun@gmail.com2025-04-15T19:08:55+08:00GitHubnoreply@github.com2025-04-15T12:08:55+01:00CANN: Add x86 build ci (#12950)
195584778e97703740d8ac5fb64e14d83b80eafa0f3c510676475f885ec064ff147af9f20ee7a9b12a50David Huang1969802+hjc4869@users.noreply.github.com2025-04-15T17:20:38+08:00GitHubnoreply@github.com2025-04-15T11:20:38+02:00CUDA/HIP: Share the same unified memory allocation logic. (#12934)
1956510676475f885ec064ff147af9f20ee7a9b12a50daa422881a0ec7944771bcc8ff8de34d11f5bd3bAkarshan Biswasakarshan@menlo.ai2025-04-15T14:07:42+05:30GitHubnoreply@github.com2025-04-15T10:37:42+02:00SYCL: Add ROPE vision kernel (#12887)
1957daa422881a0ec7944771bcc8ff8de34d11f5bd3beccc7a1602f0752507de4aaad1008b9618a282c8Juk Armstrong69222624+jukofyork@users.noreply.github.com2025-04-15T07:49:57+01:00GitHubnoreply@github.com2025-04-15T09:49:57+03:00llama : DeepSeek V2/V3 MLA implementation (#12801)
1958eccc7a1602f0752507de4aaad1008b9618a282c80019279bb56f028e4eee19b59b19750736c719f7Srihari-mcw96763064+Srihari-mcw@users.noreply.github.com2025-04-15T11:52:36+05:30GitHubnoreply@github.com2025-04-15T09:22:36+03:00ggml : Add AVX512 implementation of GEMM - Q4_Kx8 (#12829)
19590019279bb56f028e4eee19b59b19750736c719f7b0c75ac9f93322b45e3766e149417334b4fd1ed9Chenguang Li757486878@qq.com2025-04-15T10:09:35+08:00GitHubnoreply@github.com2025-04-15T10:09:35+08:00CANN: Opt ROPE optimization (#12865)
1960b0c75ac9f93322b45e3766e149417334b4fd1ed9d6d2c2ab8c8865784ba9fef37f2b2de3f2134d33Xinpeng Dou15529241576@163.com2025-04-15T10:04:24+08:00GitHubnoreply@github.com2025-04-15T10:04:24+08:00CANN: Optimize CANN buffer pool memory management (#12875)
1961d6d2c2ab8c8865784ba9fef37f2b2de3f2134d3375afa0ae31f0a51aaadcc5ff146eb7a32a7f9088Russyyds161207317+Russyyds@users.noreply.github.com2025-04-15T01:18:20+08:00GitHubnoreply@github.com2025-04-14T19:18:20+02:00Add performance print for gemma3 in example (#12929)
196275afa0ae31f0a51aaadcc5ff146eb7a32a7f9088c772d549264c1be058411312a54049e0dc86a037Akarshan Biswasakarshan@menlo.ai2025-04-14T17:53:53+05:30GitHubnoreply@github.com2025-04-14T14:23:53+02:00SYCL: Fix im2col (#12910)
1963c772d549264c1be058411312a54049e0dc86a03781c7e64fc239288e91a58adad9145110e0353822Radoslav Gerganovrgerganov@gmail.com2025-04-14T13:59:34+03:00GitHubnoreply@github.com2025-04-14T13:59:34+03:00rpc : use ggml_context_ptr (#12938)
196481c7e64fc239288e91a58adad9145110e0353822526739b879c9d6a702ed681138611197fa4d3f18Neo Zhang Jianyujianyu.zhang@intel.com2025-04-14T18:19:07+08:00GitHubnoreply@github.com2025-04-14T18:19:07+08:00dsiable curl lib check, this action is missed by commit bd3f59f81289b920bcc597a208c14f55e39ed37e (#12761) (#12937)
1965526739b879c9d6a702ed681138611197fa4d3f18a25355e2643b1feb2fcbbc4c00312aa86370d858Georgi Gerganovggerganov@gmail.com2025-04-14T08:52:10+03:00Georgi Gerganovggerganov@gmail.com2025-04-14T09:26:15+03:00sync : ggml
1966a25355e2643b1feb2fcbbc4c00312aa86370d858e959d32b1c16c02c0bc0bcdc7bc3f7077bc32f99cmdr2secondary.cmdr2@gmail.com2025-04-11T12:14:19+05:30Georgi Gerganovggerganov@gmail.com2025-04-14T09:26:15+03:00cpu: fix cpu backend's supports-op for GET_ROWS_BACK. fixes a fatal when running test-backend-ops with only the CPU backend (ggml/1190)
1967e959d32b1c16c02c0bc0bcdc7bc3f7077bc32f99307bfa253dea07c9270e78fa53b133504e9c3c9dSXXsxx1136965276@gmail.com2025-04-14T13:47:55+08:00GitHubnoreply@github.com2025-04-14T08:47:55+03:00ggml: use _mm[512/256]_dpbusd[_avx]_epi32 to directly accumulate into the result register (#12773)
1968307bfa253dea07c9270e78fa53b133504e9c3c9d71e90e8813f90097701e62f7fce137d96ddf41e2Alan Grayagray3@users.noreply.github.com2025-04-13T22:12:21+01:00GitHubnoreply@github.com2025-04-13T23:12:21+02:00ggml: disable CUDA graphs for unsupported DUP and CONT node types (#12891)
196971e90e8813f90097701e62f7fce137d96ddf41e2bc091a4dc585af25c438c8473285a8cfec5c7695Ed Addario29247825+EAddario@users.noreply.github.com2025-04-13T19:29:28+01:00GitHubnoreply@github.com2025-04-13T21:29:28+03:00quantize: Handle user-defined quantization levels for additional tensors (#12511)
1970bc091a4dc585af25c438c8473285a8cfec5c7695a4837577aae59c0ae640f3810094724bcac6bb28Prajwal B Mehendarkarprajwal.b.mehendarkar@ibm.com2025-04-12T21:03:39+05:30GitHubnoreply@github.com2025-04-12T17:33:39+02:00common : Define cache directory on AIX (#12915)
1971a4837577aae59c0ae640f3810094724bcac6bb28e59ea539b83d2c7947c99bd350549364dbba450cJeff Bolzjbolz@nvidia.com2025-04-12T03:44:48-05:00GitHubnoreply@github.com2025-04-12T10:44:48+02:00vulkan: use aligned loads for flash attention mask (#12853)
1972e59ea539b83d2c7947c99bd350549364dbba450cc94085df2871d2cad11f6411304d7798d7dd4cddMatt Clayton156335168+mattjcly@users.noreply.github.com2025-04-12T01:29:03-04:00GitHubnoreply@github.com2025-04-12T07:29:03+02:00llava: Fix cpu-only clip image encoding sefault (#12907)
1973c94085df2871d2cad11f6411304d7798d7dd4cdde8a62631b3b05bcf2ad0e0c686881a9f3e3f03caGeorgi Gerganovggerganov@gmail.com2025-04-11T23:37:41+03:00GitHubnoreply@github.com2025-04-11T23:37:41+03:00server : add VSCode's Github Copilot Chat support (#12896)
1974e8a62631b3b05bcf2ad0e0c686881a9f3e3f03cab6930ebc421e20399663bbd3bc7b7266d5236d06yuri@FreeBSDyurivict@users.noreply.github.com2025-04-11T13:04:14-07:00GitHubnoreply@github.com2025-04-11T22:04:14+02:00rpc : Set cache directory in rpc-server.cpp on FreeBSD (#12903)
1975b6930ebc421e20399663bbd3bc7b7266d5236d0668b08f36d047bfa048ebd7d16262c53bf9ece701Olivier Chafikochafik@users.noreply.github.com2025-04-11T12:47:52-07:00GitHubnoreply@github.com2025-04-11T21:47:52+02:00`tool-call`: fix non-tool-calling grammar crashes w/ Qwen / Hermes 2 templates (#12900)
197668b08f36d047bfa048ebd7d16262c53bf9ece701578754b3157d662c2fdd51eaa62b6c1f43d3172cyuri@FreeBSDyurivict@users.noreply.github.com2025-04-11T12:45:44-07:00GitHubnoreply@github.com2025-04-11T21:45:44+02:00common : Define cache directory on FreeBSD (#12892)
1977578754b3157d662c2fdd51eaa62b6c1f43d3172cb2034c2b55b36b2192bdefb3b295db2a911370f5Ewan Crawfordewan.cr@gmail.com2025-04-11T15:32:14+02:00GitHubnoreply@github.com2025-04-11T15:32:14+02:00sycl: Support sycl_ext_oneapi_limited_graph (#12873)
1978b2034c2b55b36b2192bdefb3b295db2a911370f506bb53ad9b6e6d92b6ab6979927530080b1c990ctastelikefeet58414341+tastelikefeet@users.noreply.github.com2025-04-11T20:01:56+08:00GitHubnoreply@github.com2025-04-11T14:01:56+02:00contrib: support modelscope community (#12664)
197906bb53ad9b6e6d92b6ab6979927530080b1c990c0c509239445088f21265580b86733c5b184261d9Yuxuan Zhang2448370773@qq.com2025-04-11T18:10:10+08:00GitHubnoreply@github.com2025-04-11T12:10:10+02:00llama-model : add Glm4Model implementation for GLM-4-0414 (#12867)
19800c509239445088f21265580b86733c5b184261d9fccf9cae83e6c6cd31a0ecb403d237638e427d0aXuan-Son Nguyenson@huggingface.co2025-04-11T12:09:39+02:00GitHubnoreply@github.com2025-04-11T12:09:39+02:00clip : use smart pointer (⚠️ breaking change) (#12869)
1981fccf9cae83e6c6cd31a0ecb403d237638e427d0aec6c09d0fac1f2699c3ea1994e10482cb4b95e0fAkarshan Biswasakarshan@menlo.ai2025-04-11T13:33:50+05:30GitHubnoreply@github.com2025-04-11T16:03:50+08:00SYCL: Add fp16 type support to unary op kernels (#12788)
1982ec6c09d0fac1f2699c3ea1994e10482cb4b95e0f8ac9f5d765f2b1b7f821873618c0cff68ca59bc8Daniel Handanielhanchen@gmail.com2025-04-11T00:49:09-07:00GitHubnoreply@github.com2025-04-11T09:49:09+02:00convert : Llama4 RoPE fix (#12889)
19838ac9f5d765f2b1b7f821873618c0cff68ca59bc812e9158f25cb47e97ca9b8083e1ec7415f262260R0CKSTARxiaodong.ye@mthreads.com2025-04-11T15:26:17+08:00GitHubnoreply@github.com2025-04-11T09:26:17+02:00ci : Replace freediskspace to free_disk_space in docker.yml (#12861)
198412e9158f25cb47e97ca9b8083e1ec7415f2622605b1f13cb64dbb0de7e95dae86725928d350c188cDaniel Beveniusdaniel.bevenius@gmail.com2025-04-11T09:24:34+02:00GitHubnoreply@github.com2025-04-11T09:24:34+02:00xcf : add check for visionos build version (#12854)
19855b1f13cb64dbb0de7e95dae86725928d350c188c8b91d5355a84bbb5a73fd23fb658272473240efeXuan-Son Nguyenson@huggingface.co2025-04-11T09:23:37+02:00GitHubnoreply@github.com2025-04-11T09:23:37+02:00convert : proper tensor name mapping for llama4 (#12870)
19868b91d5355a84bbb5a73fd23fb658272473240efe0fed24c34787d2aa916ed204db88889591ad6e55Xuan-Son Nguyenson@huggingface.co2025-04-11T08:49:50+02:00GitHubnoreply@github.com2025-04-11T08:49:50+02:00llama : correct rms norm for llama 4 (#12882)
19870fed24c34787d2aa916ed204db88889591ad6e5547ba87d0a4f91eb181ea60f8a7fc2539b123aeafAaron Teo57927438+taronaeo@users.noreply.github.com2025-04-11T13:20:07+08:00GitHubnoreply@github.com2025-04-11T08:20:07+03:00ggml: fix compilation error s390x (#12848)
198847ba87d0a4f91eb181ea60f8a7fc2539b123aeaf1d2b613445bab6b179296aa63c8ee346fb947cc4Georgi Gerganovggerganov@gmail.com2025-04-11T00:08:23+03:00Georgi Gerganovggerganov@gmail.com2025-04-11T00:17:47+03:00sync : ggml
19891d2b613445bab6b179296aa63c8ee346fb947cc4eb420e11484ef32cc1abedb2a26ecef1bbf1e31bGeorgi Gerganovggerganov@gmail.com2025-04-11T00:04:25+03:00Georgi Gerganovggerganov@gmail.com2025-04-11T00:17:47+03:00tests : fix init order (#0)
1990eb420e11484ef32cc1abedb2a26ecef1bbf1e31bcb79c2e7fa28e874694c14598c1fd2fde82263e1Georgi Gerganovggerganov@gmail.com2025-04-10T23:59:16+03:00Georgi Gerganovggerganov@gmail.com2025-04-11T00:17:47+03:00sync : ggml
1991cb79c2e7fa28e874694c14598c1fd2fde82263e1fe92821ea9ae53f3088cf2699a9e102448295fa0cmdr2secondary.cmdr2@gmail.com2025-04-10T17:53:08+05:30Georgi Gerganovggerganov@gmail.com2025-04-11T00:17:47+03:00ggml: don't include arm_neon.h when using CUDA 12 with ARM Neon (ggml/1187)
1992fe92821ea9ae53f3088cf2699a9e102448295fa0459895c32629e36cc3ca53bf3596d3e1322cfe09Diego Devesaslarengh@gmail.com2025-04-09T12:32:13+02:00Georgi Gerganovggerganov@gmail.com2025-04-11T00:17:47+03:00ggml : add bilinear upscale support (ggml/1185)
1993459895c32629e36cc3ca53bf3596d3e1322cfe09e4bf72d631434b38656cb97e2411826d900433e8Diego Devesaslarengh@gmail.com2025-04-09T12:31:34+02:00Georgi Gerganovggerganov@gmail.com2025-04-11T00:17:47+03:00ggml : add more generic custom op, remove deprecated custom ops (ggml/1183)
1994e4bf72d631434b38656cb97e2411826d900433e88b9cc7cdd8a0dcf0176c60c755322c95b5965299Georgi Gerganovggerganov@gmail.com2025-04-10T23:59:01+03:00Georgi Gerganovggerganov@gmail.com2025-04-11T00:17:47+03:00scripts : fix sync-ggml-am.sh
19958b9cc7cdd8a0dcf0176c60c755322c95b596529964eda5deb9859e87a020e56bab5d2f9ca956f1deXuan-Son Nguyenson@huggingface.co2025-04-10T22:57:16+02:00GitHubnoreply@github.com2025-04-10T22:57:16+02:00llava : introduce libmtmd (#12849)
199664eda5deb9859e87a020e56bab5d2f9ca956f1defe5b78c89670b2f37ecb216306bed3e677b49d9fXuan-Son Nguyenson@huggingface.co2025-04-10T17:24:44+02:00GitHubnoreply@github.com2025-04-10T17:24:44+02:00convert : ability to lazy-load safetensors remotely without downloading to disk (#12820)
1997fe5b78c89670b2f37ecb216306bed3e677b49d9f11d07e1e69138b46375e9267b31acd58e3813577Chenguang Li757486878@qq.com2025-04-10T08:51:52+08:00GitHubnoreply@github.com2025-04-10T08:51:52+08:00CANN: Support more ops (#12841)
199811d07e1e69138b46375e9267b31acd58e3813577b0091ecc1e5c0f689be856fade3803a534f35c9fPrajwal B Mehendarkarprajwal.b.mehendarkar@ibm.com2025-04-10T04:48:01+05:30GitHubnoreply@github.com2025-04-10T01:18:01+02:00Fixes #12823 (#12830)
1999b0091ecc1e5c0f689be856fade3803a534f35c9f31f7803bc4e7c0dcc279ee04c2ecfb76b2afdd3eRudi Servorudiservo@gmail.com2025-04-09T23:17:12ZGitHubnoreply@github.com2025-04-10T01:17:12+02:00docker : added all CPU to GPU images (#12749)
200031f7803bc4e7c0dcc279ee04c2ecfb76b2afdd3e2391506ace6abb56186def40c7107fdfa694ed55Piotr Kubajpkubaj@anongoth.pl2025-04-09T23:00:34ZGitHubnoreply@github.com2025-04-10T01:00:34+02:00ggml-cpu-impl.h: do not redefine bool on POWER9 (#12856)
20012391506ace6abb56186def40c7107fdfa694ed55d3bd7193ba66c15963fd1c59448f22019a8caf6ePiotr Kubajpkubaj@anongoth.pl2025-04-09T23:00:25ZGitHubnoreply@github.com2025-04-10T01:00:25+02:00ggml-impl.h: fix build on POWER9 (#12855)
2002d3bd7193ba66c15963fd1c59448f22019a8caf6ed9a63b2f2e91cdcb0eda211b7f49fadcdea0f664Bo Zheng368586905@qq.com2025-04-09T17:47:36+08:00GitHubnoreply@github.com2025-04-09T11:47:36+02:00llama : Support Qwen3 and Qwen3MoE (#12828)
2003d9a63b2f2e91cdcb0eda211b7f49fadcdea0f6648ed71242f464dc0a3fb3cffcfe064e55bdec72f9R0CKSTARxiaodong.ye@mthreads.com2025-04-09T17:22:30+08:00GitHubnoreply@github.com2025-04-09T11:22:30+02:00musa: enable freediskspace for docker image build (#12839)
20048ed71242f464dc0a3fb3cffcfe064e55bdec72f9381603a77504ad1788965f694094540c1bed9ea2Romain Biessyromain.biessy@codeplay.com2025-04-09T11:22:04+02:00GitHubnoreply@github.com2025-04-09T11:22:04+02:00sycl: update documentation to use -no-cnv (#12845)
2005381603a77504ad1788965f694094540c1bed9ea265a69e6e1b6d55cd5f78f8bcdfaba8a8c59a8d96Plamen Minevpacominev@gmail.com2025-04-09T11:11:11+03:00GitHubnoreply@github.com2025-04-09T10:11:11+02:00ci: detach common from the library (#12827)
200665a69e6e1b6d55cd5f78f8bcdfaba8a8c59a8d9647277d6d1d0d515cff34292a1a78a0d1b7252350Xuan-Son Nguyenson@huggingface.co2025-04-09T10:09:53+02:00GitHubnoreply@github.com2025-04-09T10:09:53+02:00clip : do not print ftype (#12832)
200747277d6d1d0d515cff34292a1a78a0d1b72523506e1c4cebdb697f925c523d3a969128d945161bddGeorgi Gerganovggerganov@gmail.com2025-04-09T10:54:42+03:00GitHubnoreply@github.com2025-04-09T10:54:42+03:00readme : add rpc backend (#12842)
20086e1c4cebdb697f925c523d3a969128d945161bdd0090950f679475c5ecaac2f7bca5049cca96492bChenguang Li757486878@qq.com2025-04-09T14:04:14+08:00GitHubnoreply@github.com2025-04-09T14:04:14+08:00CANN: Support Opt CONV_TRANSPOSE_1D and ELU (#12786)
20090090950f679475c5ecaac2f7bca5049cca96492b7ecd780b1a1d5214b8d04c25ebfc194d310816edJeff Bolzjbolz@nvidia.com2025-04-09T00:25:08-05:00GitHubnoreply@github.com2025-04-09T07:25:08+02:00vulkan: In coopmat2 mmq, load q4_k/q5_k scales through shared memory (#12833)
20107ecd780b1a1d5214b8d04c25ebfc194d310816ed7538246e7ce0606694c38055cc2fc9f60535be6cJeff Bolzjbolz@nvidia.com2025-04-09T00:12:57-05:00GitHubnoreply@github.com2025-04-09T07:12:57+02:00vulkan: Use fp16 for the flash attention P*V multiplication (#12783)
20117538246e7ce0606694c38055cc2fc9f60535be6cb32efad2bc42460637c3a364c9554ea8217b3d7fSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-04-08T23:21:31+02:00GitHubnoreply@github.com2025-04-08T23:21:31+02:00cuda : add f32 to bf16 copy op (#12806)
2012b32efad2bc42460637c3a364c9554ea8217b3d7fa19b5cef16d885c44c635da4a5c97113c1577de8Matt Clayton156335168+mattjcly@users.noreply.github.com2025-04-08T16:01:58-04:00GitHubnoreply@github.com2025-04-08T22:01:58+02:00llava: improve clip_ctx destructor to not memleak load_image_size (#12834)
2013a19b5cef16d885c44c635da4a5c97113c1577de878a1ba0a4f2bfed5b8b8e312592143d22e531698Georgi Gerganovggerganov@gmail.com2025-04-08T19:54:51+03:00GitHubnoreply@github.com2025-04-08T19:54:51+03:00llama : fix FA when KV cache is not used (i.e. embeddings) (#12825)
201478a1ba0a4f2bfed5b8b8e312592143d22e5316982dabf759e7c8c827d38cdd18d0792070e6a4f4e1Xuan-Son Nguyenson@huggingface.co2025-04-08T18:37:06+02:00GitHubnoreply@github.com2025-04-08T18:37:06+02:00server : fix thread.join() on exit (#12831)
20152dabf759e7c8c827d38cdd18d0792070e6a4f4e11d343b4069c74b2c7b19ae84260cd98aa2320a9adm4sunrisedm4@gmail.com2025-04-08T21:49:13+08:00GitHubnoreply@github.com2025-04-08T15:49:13+02:00llava: add more helper functions to check projector types in clip context (#12824)
20161d343b4069c74b2c7b19ae84260cd98aa2320a9a8ca6e1c3a4deb6bb27ee294bfd5706098d94ae88Prajwal B Mehendarkarprajwal.b.mehendarkar@ibm.com2025-04-08T18:00:59+05:30GitHubnoreply@github.com2025-04-08T14:30:59+02:00arg : Including limits file on AIX (#12822)
20178ca6e1c3a4deb6bb27ee294bfd5706098d94ae88656babd6c21a3b9b3622324cfcc80a2ab78da25bcharacharm123120856+characharm@users.noreply.github.com2025-04-08T14:14:59+05:00GitHubnoreply@github.com2025-04-08T11:14:59+02:00server : webui : Improve Chat Input with Auto-Sizing Textarea (#12785)
2018a226bc7a9ac50551f9f113808de0f0046837f1881466621e738779eefe1bb672e17dc55d63d166bbcompiladegit@compilade.net2025-04-08T03:03:07-04:00GitHubnoreply@github.com2025-04-08T09:03:07+02:00gguf-py : support lazy tensor splitting (#12809)
20191466621e738779eefe1bb672e17dc55d63d166bb82974011f312057b446c27267105bd7ad3810599Xuan-Son Nguyenson@huggingface.co2025-04-07T23:06:44+02:00GitHubnoreply@github.com2025-04-07T23:06:44+02:00llama : Support llama 4 text-only (#12791)
202082974011f312057b446c27267105bd7ad38105994ccea213bc629c4eef7b520f7f6c59ce9bbdaca0lhezquic_lih@quicinc.com2025-04-07T13:22:54-07:00GitHubnoreply@github.com2025-04-07T13:22:54-07:00opencl: better identify Adreno GPU (#12760)
20214ccea213bc629c4eef7b520f7f6c59ce9bbdaca01a1ab7e7a4a9b6e6440c1d9965d2b9d1b7e7dafbstduhpfstephduh@live.fr2025-04-07T17:47:08+02:00GitHubnoreply@github.com2025-04-07T18:47:08+03:00hellaswag: display estimated score confidence interval (#12797)
20221a1ab7e7a4a9b6e6440c1d9965d2b9d1b7e7dafba4e46e28f99d7f64f38d8328cdb6bee5a3a1cd03Georgi Gerganovggerganov@gmail.com2025-04-07T13:18:07+03:00Georgi Gerganovggerganov@gmail.com2025-04-07T18:44:17+03:00cuda : fix HIP and MUSA BF16 (#0)
2023a4e46e28f99d7f64f38d8328cdb6bee5a3a1cd03ff067dbcb9e6ca4ed464d3db999ff8e9c503498bGeorgi Gerganovggerganov@gmail.com2025-04-07T12:32:39+03:00Georgi Gerganovggerganov@gmail.com2025-04-07T18:44:17+03:00sync : ggml
2024ff067dbcb9e6ca4ed464d3db999ff8e9c503498b36ca8b362885e4b4984d72e348ba403568864a95Georgi Gerganovggerganov@gmail.com2025-04-07T12:25:15+03:00Georgi Gerganovggerganov@gmail.com2025-04-07T18:44:17+03:00ggml : simplify Arm fp16 CPU logic (ggml/1177)
202536ca8b362885e4b4984d72e348ba403568864a95995083e4ed24933e6a289472f9de0f0b53ca5ecaSigbjørn Skjæretsigbjorn.skjaeret@scala.com2025-04-04T21:05:12+02:00Georgi Gerganovggerganov@gmail.com2025-04-07T18:44:17+03:00CUDA: don't convert BF16 weights to FP32 (ggml/1174)
2026995083e4ed24933e6a289472f9de0f0b53ca5eca518a01480eb3a7c80a4951b430db9dee55428310cmdr2secondary.cmdr2@gmail.com2025-04-02T17:46:16+05:30Georgi Gerganovggerganov@gmail.com2025-04-07T18:44:17+03:00cpu: move all the operators into a separate c++ file (except mul_mat) (ggml/1167)
2027518a01480eb3a7c80a4951b430db9dee55428310e391d3ee8ddae86be70c034de1082ad51c55e211zhouwgzhouwg2000@gmail.com2025-04-07T23:22:57+08:00GitHubnoreply@github.com2025-04-07T17:22:57+02:00sycl: remove redundant memcopy in function ggml_backend_sycl_buffer_set_tensor (#12734)
2028e391d3ee8ddae86be70c034de1082ad51c55e211bd3f59f81289b920bcc597a208c14f55e39ed37eXuan-Son Nguyenson@huggingface.co2025-04-07T14:37:28+02:00GitHubnoreply@github.com2025-04-07T15:37:28+03:00ci : no curl on ggml-ci (#12796)
2029bd3f59f81289b920bcc597a208c14f55e39ed37e52b3d71f128c1eeab39b918adf1f7a8f5e256619Xuan-Son Nguyenson@huggingface.co2025-04-07T13:35:19+02:00GitHubnoreply@github.com2025-04-07T13:35:19+02:00cmake : enable curl by default (#12761)
203052b3d71f128c1eeab39b918adf1f7a8f5e256619d0d5b2232b3826cac2c6e62d0244a474ba79860fzhouwgzhouwg2000@gmail.com2025-04-07T19:34:14+08:00GitHubnoreply@github.com2025-04-07T19:34:14+08:00CANN: fix typo in ggml-cann (#12733)
2031d0d5b2232b3826cac2c6e62d0244a474ba79860f916c83bfe7f8b08ada609c3b8e583cf5301e594bhipuddinghuafengchun@gmail.com2025-04-07T17:10:36+08:00GitHubnoreply@github.com2025-04-07T17:10:36+08:00CANN: Refactor to reduce duplicate code (#12731)
2032916c83bfe7f8b08ada609c3b8e583cf5301e594b0c74b04376b0b9efc096480fe10f866afc8d7c1cR0CKSTARxiaodong.ye@mthreads.com2025-04-06T21:23:54+08:00GitHubnoreply@github.com2025-04-06T15:23:54+02:00musa: fix compilation warnings in mp_22/31 (#12780)
20330c74b04376b0b9efc096480fe10f866afc8d7c1c80b717d493a9a5bae7167ad2384c12c60bb2ef20Jeff Bolzjbolz@nvidia.com2025-04-06T04:03:47-05:00GitHubnoreply@github.com2025-04-06T11:03:47+02:00vulkan: fix NaN issue in flash attention shader (#12776)
203480b717d493a9a5bae7167ad2384c12c60bb2ef206bf28f0111ff9f21b3c1b1eace20c590281e7ba6Jeff Bolzjbolz@nvidia.com2025-04-06T03:47:13-05:00GitHubnoreply@github.com2025-04-06T10:47:13+02:00vulkan: Use unclamped loads for flash attention mask (#12720)
20356bf28f0111ff9f21b3c1b1eace20c590281e7ba6f1e3eb4249db68d97be352c0adf16eef7ae537950cc4mpicard12@live.de2025-04-05T18:04:03+02:00GitHubnoreply@github.com2025-04-05T18:04:03+02:00Vulkan: Tune Vulkan mmq int dot shader for performance (#12767)
2036f1e3eb4249db68d97be352c0adf16eef7ae537950364178ca2452bd241a4cdd4350d0d7a6ff8c4b4Sergey Fedorovvital.had@gmail.com2025-04-05T23:46:00+08:00GitHubnoreply@github.com2025-04-05T17:46:00+02:00common : fix includes in arg.cpp and gemma3-cli.cpp (#12766)
20370364178ca2452bd241a4cdd4350d0d7a6ff8c4b4c6ff5d2a8da2587cc78c9ede9171dfc3f076c757Xuan-Son Nguyenson@huggingface.co2025-04-05T17:17:40+02:00GitHubnoreply@github.com2025-04-05T17:17:40+02:00clip : refactor clip_init, add tests (#12757)
2038c6ff5d2a8da2587cc78c9ede9171dfc3f076c7577a84777f42a9b3ba47db5d20b7662f8ddf92f652エシュナヴァリシア148695646+eternaphia@users.noreply.github.com2025-04-05T21:31:42+08:00GitHubnoreply@github.com2025-04-05T15:31:42+02:00common: custom hf endpoint support (#12769)
20397a84777f42a9b3ba47db5d20b7662f8ddf92f6523e1d29348b5d77269f6931500dd1c1a729d429c8Olivier Chafikochafik@users.noreply.github.com2025-04-04T13:16:39-07:00GitHubnoreply@github.com2025-04-04T21:16:39+01:00sync: minja (#12739)
20403e1d29348b5d77269f6931500dd1c1a729d429c81be76e4620ddc99b3cbff0f206436117ae561047Georgi Gerganovggerganov@gmail.com2025-04-04T21:48:10+03:00GitHubnoreply@github.com2025-04-04T21:48:10+03:00kv-cache : simplify + fix warning for recurrent models (#12756)
20411be76e4620ddc99b3cbff0f206436117ae561047b7723942970b70412793f1926a35cfb93d5c157cbandoti141645996+bandoti@users.noreply.github.com2025-04-04T14:05:12-03:00GitHubnoreply@github.com2025-04-04T14:05:12-03:00ci: add Linux cross-compile build (#12428)
2042b7723942970b70412793f1926a35cfb93d5c157c23106f94ea2bc3da929afb7330655fd5515d08dcNauful Shaikhnauful@gmail.com2025-04-04T09:09:52-05:00GitHubnoreply@github.com2025-04-04T16:09:52+02:00server : webui : Upgrade daisyui, tailwindcss. (#12735)
204323106f94ea2bc3da929afb7330655fd5515d08dc94148ba330968bbfb8d9ecc67751bdc2218486cdnick huangnickhuang99@hotmail.com2025-04-04T22:09:12+08:00GitHubnoreply@github.com2025-04-04T16:09:12+02:00gguf-split : --merge now respects --dry-run option (#12681)
204494148ba330968bbfb8d9ecc67751bdc2218486cd9ac4d611d05b03f961e627f4f399e4377bdb4ad3Nicolò Scipionenicolo.scipione@codeplay.com2025-04-04T16:00:46+02:00GitHubnoreply@github.com2025-04-04T16:00:46+02:00sycl: allow ggml-sycl configuration and compilation using Visual Studio project/solution (#12625)
20459ac4d611d05b03f961e627f4f399e4377bdb4ad3348888e0dc469a476f9e9eccb394893c513daab8Ronny Brendelronnyb@nvidia.com2025-04-04T15:12:40+02:00GitHubnoreply@github.com2025-04-04T10:12:40-03:00cmake: fix ggml-shaders-gen compiler paths containing spaces (#12747)
2046348888e0dc469a476f9e9eccb394893c513daab874d4f5b041ad837153b0e90fc864b8290e01d8d5Daniel Beveniusdaniel.bevenius@gmail.com2025-04-04T10:24:12+02:00GitHubnoreply@github.com2025-04-04T10:24:12+02:00docs : add XCFramework section to README.md [no ci] (#12746)
204774d4f5b041ad837153b0e90fc864b8290e01d8d535e592eb30832187412360912ab8f2f5b7984df1Jeff Bolzjbolz@nvidia.com2025-04-04T00:54:35-05:00GitHubnoreply@github.com2025-04-04T07:54:35+02:00vulkan: Hybrid waitForFences/getFenceStatus to reduce fence latency (#12630)
204835e592eb30832187412360912ab8f2f5b7984df17d7b1bafa7980603a61cb102879fe38a33f66c08Jeff Bolzjbolz@nvidia.com2025-04-04T00:53:20-05:00GitHubnoreply@github.com2025-04-04T07:53:20+02:00vulkan: set cmake minimum and project name in vulkan-shaders (#12744)
20497d7b1bafa7980603a61cb102879fe38a33f66c08c262beddf29f3f3be5bbbf167b56029a19876956lhezquic_lih@quicinc.com2025-04-03T22:18:17-07:00GitHubnoreply@github.com2025-04-03T22:18:17-07:00opencl: update doc for OpenCL (#12702)
2050c262beddf29f3f3be5bbbf167b56029a198769565dd5d1ab00d074e3b7c02ca3ae12f6bf3e86336aGaurav Garg52341457+gaugarg-nv@users.noreply.github.com2025-04-03T21:50:29+05:30GitHubnoreply@github.com2025-04-03T18:20:29+02:00CUDA: Prefer vector flash decoding kernel for Gemma models (#12738)
20515dd5d1ab00d074e3b7c02ca3ae12f6bf3e86336a1c059995e080e37aeaacaae35850fc4c044b9dbeyumeyaoyumeyao@gmail.com2025-04-03T23:32:54+08:00GitHubnoreply@github.com2025-04-03T18:32:54+03:00vocab : use string_view::find() to avoid unnecessary looking up beyond the fragment range (#12706)
20521c059995e080e37aeaacaae35850fc4c044b9dbe2004644b7a5da6fe080e51861ab583480280f1d3Jeff Bolzjbolz@nvidia.com2025-04-03T10:08:26-05:00GitHubnoreply@github.com2025-04-03T10:08:26-05:00vulkan: Fix missing cmake logic for dot product extension (#12721)
20532004644b7a5da6fe080e51861ab583480280f1d35f696e88e0eb490c8028421d3c5419df9dcf5385Atharva Dubeyatharva.dubey@codeplay.com2025-04-03T13:12:39+01:00GitHubnoreply@github.com2025-04-03T15:12:39+03:00ci : add env variable in ggml-ci and document the same in SYCL.md (#12736)
20545f696e88e0eb490c8028421d3c5419df9dcf5385193c3e03a63ccda3ac3d6a2999e41e6d1414fe23R0CKSTARxiaodong.ye@mthreads.com2025-04-03T19:51:35+08:00GitHubnoreply@github.com2025-04-03T13:51:35+02:00sync : minja (inclusionAI/Ling) and update tests (#12699)
2055193c3e03a63ccda3ac3d6a2999e41e6d1414fe2365cfe136a0793b2fdf5af8bdb1ab2cf10c3e1d2ea3sh38979186+A3shTnT@users.noreply.github.com2025-04-03T15:32:55+08:00GitHubnoreply@github.com2025-04-03T09:32:55+02:00fix MUSA compiler warning (#12704)
205665cfe136a0793b2fdf5af8bdb1ab2cf10c3e1d2e3f9da22c2b21a2cef216de50006436ef1cab8764Chenguang Li757486878@qq.com2025-04-03T15:18:08+08:00GitHubnoreply@github.com2025-04-03T15:18:08+08:00CANN: Support operator SIN COS ARGMAX (#12709)
20573f9da22c2b21a2cef216de50006436ef1cab87642a0dc97e56eac6db0a4016f0b45da6d0a0055ef2Alan Grayagray3@users.noreply.github.com2025-04-03T02:31:15+01:00GitHubnoreply@github.com2025-04-03T03:31:15+02:00Simplify and improve CUDA graphs through use of indirect copy pointers (#9017)
20582a0dc97e56eac6db0a4016f0b45da6d0a0055ef297a20c012be2f9bfbeee0209405a31001f93ccf9hipuddi
The file is too large to be shown. View Raw
Reference in New Issue View Git Blame Copy Permalink