Files
calculet-npu-research-archive/history/llama.cpp/llama.cpp-recovered-unreachable-calrt-add-slice.patch
T

27 lines
997 B
Diff

commit 1d248d956f6cbcf5fdd4169d5467678ead035fe2
Author: wangbomeng <wangbomeng@calculet.tech>
AuthorDate: Tue Mar 24 02:17:22 2026 +0800
Commit: wangbomeng <wangbomeng@calculet.tech>
CommitDate: Tue Mar 24 02:17:22 2026 +0800
calrt: add slice
---
src/llama-calrt.cpp | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/src/llama-calrt.cpp b/src/llama-calrt.cpp
index 5bbdd52e..e038a4c2 100644
--- a/src/llama-calrt.cpp
+++ b/src/llama-calrt.cpp
@@ -564,6 +564,10 @@ void calrt_context::calrt_infer(calrt_seqs_info & seqs) {
// makeup_ubatch(ubatch);
copy_data_to_ibuf(*pIbuf, ubatch, model_name);
+ //only copy promt tokens
+ //pObuf->SliceTensorByName(pObuf->GetTensorByName("outputs[0]")->GetDataPtr(),"outputs[0]", 0, cur_seq_len * sizeof(ggml_bf16_t));
+ pObuf->GetTensorByName("outputs[0]")->SliceTensor(0, cur_seq_len * sizeof(ggml_bf16_t));
+
calrt::infer(vdev, model, pIbuf.get(), pObuf.get());
pObuf->Wait();