commit c44438579a1cc573cbdb2519dda46ccd04ddc817 Merge: 207ca38c f06f10a5 Author: Bomeng Wang AuthorDate: Thu Apr 9 10:37:03 2026 +0800 Commit: Bomeng Wang CommitDate: Thu Apr 9 10:37:03 2026 +0800 WIP on dev: 207ca38c delet extra printf tools/server/server.cpp | 22 +++++++++++++--------- 1 file changed, 13 insertions(+), 9 deletions(-) diff --cc tools/server/server.cpp index 868d17b4,868d17b4..d2a035ba --- a/tools/server/server.cpp +++ b/tools/server/server.cpp @@@ -1875,7 -1875,7 +1875,7 @@@ struct server_slot GGML_ASSERT(task); auto previous_msg = chat_msg; -- SRV_DBG("Parsing chat message: %s\n", generated_text.c_str()); ++ //SRV_DBG("Parsing chat message: %s\n", generated_text.c_str()); auto new_msg = common_chat_parse( generated_text, /* is_partial= */ stop != STOP_TYPE_EOS, @@@ -1920,13 -1920,13 +1920,17 @@@ } void print_timings() const { -- const double t_prompt = t_prompt_processing / n_prompt_tokens_processed; -- const double n_prompt_second = 1e3 / t_prompt_processing * n_prompt_tokens_processed; ++ const double t_copyout_total = t_inference_outcopy / 32 * 31; ++ const double t_copyout_decode = (t_inference_outcopy - t_prefill_inference_outcopy) / 32 * 31; ++ const double t_copyout_prefill = t_prefill_inference_outcopy / 32 * 31; -- const double t_gen = t_token_generation / n_decoded; -- const double n_gen_second = 1e3 / t_token_generation * n_decoded; ++ const double t_prompt = (t_prompt_processing - t_copyout_prefill) / n_prompt_tokens_processed; ++ const double n_prompt_second = 1e3 / (t_prompt_processing - t_copyout_prefill) * n_prompt_tokens_processed; -- const double t_total = t_prompt_processing + t_token_generation; ++ const double t_gen = (t_token_generation - t_copyout_decode) / n_decoded; ++ const double n_gen_second = 1e3 / (t_token_generation - t_copyout_decode) * n_decoded; ++ ++ const double t_total = t_prompt_processing + t_token_generation -t_copyout_total ; const double t_framework_other = t_total - t_inference_accumulated - t_inference_incopy - t_inference_outcopy - t_sampling_accumulated - t_unknown; const double t_prompt_inference = t_prefill_inference_accumulated / n_prompt_tokens_processed; @@@ -1950,9 -1950,9 +1954,9 @@@ " [Device] Unknown : %10.2f ms \n" " [CPU] Sampling: %10.2f ms \n" " [CPU] Fw Ovhd : %10.2f ms (Batch Prep, Logic, etc.)\n", -- t_prompt_processing, n_prompt_tokens_processed, t_prompt, n_prompt_second, -- t_token_generation, n_decoded, t_gen, n_gen_second, -- t_prompt_processing + t_token_generation, n_prompt_tokens_processed + n_decoded, ++ t_prompt_processing - t_copyout_prefill , n_prompt_tokens_processed, t_prompt, n_prompt_second, ++ t_token_generation - t_copyout_decode , n_decoded, t_gen, n_gen_second, ++ t_prompt_processing + t_token_generation - t_copyout_total, n_prompt_tokens_processed + n_decoded, t_inference_accumulated, t_prefill_inference_accumulated, t_prompt_inference, t_inference_accumulated - t_prefill_inference_accumulated, t_gen_inference, t_inference_incopy, t_prefill_inference_incopy, t_prompt_incopy, t_inference_incopy - t_prefill_inference_incopy, t_gen_incopy, t_inference_outcopy, t_prefill_inference_outcopy, t_prompt_outcopy, t_inference_outcopy - t_prefill_inference_outcopy, t_gen_outcopy, commit 2a67be2e31c4b7b28fa3937893ec9e86bf5d9f4d Merge: 99dd308a 294f5e41 Author: Bomeng Wang AuthorDate: Mon Apr 27 11:41:25 2026 +0800 Commit: Bomeng Wang CommitDate: Mon Apr 27 11:41:25 2026 +0800 WIP on (no branch): 99dd308a server: fix max_seq_len common/common.cpp | 90 ++++++++++++++++++++++++++--------------------------- src/llama-calrt.cpp | 14 +++++++-- 2 files changed, 57 insertions(+), 47 deletions(-) diff --cc common/common.cpp index 625b369b,625b369b..898d4452 --- a/common/common.cpp +++ b/common/common.cpp @@@ -1058,51 -1058,51 +1058,51 @@@ struct common_init_result common_init_f params.sampling.dry_penalty_last_n = llama_n_ctx(lctx); } -- if (params.warmup) { -- LOG_WRN("%s: warming up the model with an empty run - please wait ... (--no-warmup to disable)\n", __func__); -- -- llama_set_warmup(lctx, true); -- -- std::vector tmp; -- llama_token bos = llama_vocab_bos(vocab); -- llama_token eos = llama_vocab_eos(vocab); -- -- // some models (e.g. T5) don't have a BOS token -- if (bos != LLAMA_TOKEN_NULL) { -- tmp.push_back(bos); -- } -- if (eos != LLAMA_TOKEN_NULL) { -- tmp.push_back(eos); -- } -- if (tmp.empty()) { -- tmp.push_back(0); -- } -- -- if (llama_model_has_encoder(model)) { -- #ifdef USE_CALRT -- //calrt_encode(cal_ctx, llama_batch_get_one(tmp.data(), tmp.size())); -- #else -- llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size())); -- #endif -- llama_token decoder_start_token_id = llama_model_decoder_start_token(model); -- if (decoder_start_token_id == LLAMA_TOKEN_NULL) { -- decoder_start_token_id = bos; -- } -- tmp.clear(); -- tmp.push_back(decoder_start_token_id); -- } -- if (llama_model_has_decoder(model)) { -- #ifdef USE_CALRT -- calrt_decode(cal_ctx, llama_batch_get_one(tmp.data(), std::min(tmp.size(), (size_t) params.n_batch))); -- #else -- llama_decode(lctx, llama_batch_get_one(tmp.data(), std::min(tmp.size(), (size_t) params.n_batch))); -- #endif -- } -- llama_memory_clear(llama_get_memory(lctx), true); -- llama_synchronize(lctx); -- llama_perf_context_reset(lctx); -- llama_set_warmup(lctx, false); -- } ++ // if (params.warmup) { ++ // LOG_WRN("%s: warming up the model with an empty run - please wait ... (--no-warmup to disable)\n", __func__); ++ ++ // llama_set_warmup(lctx, true); ++ ++ // std::vector tmp; ++ // llama_token bos = llama_vocab_bos(vocab); ++ // llama_token eos = llama_vocab_eos(vocab); ++ ++ // // some models (e.g. T5) don't have a BOS token ++ // if (bos != LLAMA_TOKEN_NULL) { ++ // tmp.push_back(bos); ++ // } ++ // if (eos != LLAMA_TOKEN_NULL) { ++ // tmp.push_back(eos); ++ // } ++ // if (tmp.empty()) { ++ // tmp.push_back(0); ++ // } ++ ++ // if (llama_model_has_encoder(model)) { ++ // #ifdef USE_CALRT ++ // //calrt_encode(cal_ctx, llama_batch_get_one(tmp.data(), tmp.size())); ++ // #else ++ // llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size())); ++ // #endif ++ // llama_token decoder_start_token_id = llama_model_decoder_start_token(model); ++ // if (decoder_start_token_id == LLAMA_TOKEN_NULL) { ++ // decoder_start_token_id = bos; ++ // } ++ // tmp.clear(); ++ // tmp.push_back(decoder_start_token_id); ++ // } ++ // if (llama_model_has_decoder(model)) { ++ // #ifdef USE_CALRT ++ // calrt_decode(cal_ctx, llama_batch_get_one(tmp.data(), std::min(tmp.size(), (size_t) params.n_batch))); ++ // #else ++ // llama_decode(lctx, llama_batch_get_one(tmp.data(), std::min(tmp.size(), (size_t) params.n_batch))); ++ // #endif ++ // } ++ // llama_memory_clear(llama_get_memory(lctx), true); ++ // llama_synchronize(lctx); ++ // llama_perf_context_reset(lctx); ++ // llama_set_warmup(lctx, false); ++ // } iparams.model.reset(model); iparams.context.reset(lctx); diff --cc src/llama-calrt.cpp index 9f41c955,9f41c955..e8dc07c6 --- a/src/llama-calrt.cpp +++ b/src/llama-calrt.cpp @@@ -43,6 -43,6 +43,16 @@@ std::unique_ptr i throw std::runtime_error("failed to create vdevice"); } ++ uint64_t commitId = 0; ++ uint32_t rdata = 0; ++ vDev->GetDevice()->ReadReg(0x0a238254, rdata); ++ commitId |=(((uint64_t)rdata) << 32); ++ vDev->GetDevice()->ReadReg(0x0a238258, rdata); ++ commitId |=((uint64_t)rdata); ++ ++ LLAMA_LOG_INFO("CalcoreRT: =================== CalcoreRT Version: %016llx ====================\n", commitId); ++ LLAMA_LOG_INFO("CalRT: =================== CalRT Version %s ====================\n", calrt::calrt_version()); ++ return vDev; } @@@ -664,7 -664,7 +674,7 @@@ calrt::CalbinModel * calrt_context::get } if (smodelName.empty()) { -- LLAMA_LOG_ERROR("submodel who's name contains %s not found", name.c_str()); ++ //LLAMA_LOG_ERROR("submodel who's name contains %s not found", name.c_str()); return nullptr; } @@@ -681,7 -681,7 +691,7 @@@ calrt::CalbinModel * calrt_context::get } if (smodelName.empty()) { -- LLAMA_LOG_ERROR("submodel who's name contains %s not found", name1.c_str(), name2.c_str()); ++ //LLAMA_LOG_ERROR("submodel who's name contains %s not found", name1.c_str(), name2.c_str()); return nullptr; } commit 81d4fd2c584e6df9199181e44b77fc3d939c0a67 Merge: efa1876b 499f6ff4 Author: Bomeng Wang AuthorDate: Thu Apr 30 16:18:01 2026 +0800 Commit: Bomeng Wang CommitDate: Thu Apr 30 16:18:01 2026 +0800 WIP on (no branch): efa1876b update version print src/llama-calrt.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --cc src/llama-calrt.cpp index 0dd65366,0dd65366..845d1df3 --- a/src/llama-calrt.cpp +++ b/src/llama-calrt.cpp @@@ -133,7 -133,7 +133,7 @@@ int calrt_context::decode(const llama_b // require that all tokens are output if (n_outputs_all != n_tokens_all) { LLAMA_LOG_ERROR( -- "%s: pooled embedding requires that all tokens are output (n_outputs_all = %d, n_tokens_all = %d)\n", ++ "%s: pooled embedding requires that all tokens are eoutput (n_outputs_all = %d, n_tokens_all = %d)\n", __func__, n_outputs_all, n_tokens_all); return -1; }