vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix the Bug: inference running result is garbled in debug running model for LM models who's type is Q4_0 class

Browse source
This commit is contained in:
leo-pony 2024-10-21 19:13:21 +08:00
parent 1657447b2b
commit 4d4ae1c9a1
1 changed files with 0 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

14
ggml/src/ggml-cann.cpp
View file

@ -851,13 +851,6 @@ static void ggml_backend_cann_buffer_set_tensor(
void *transform_buffer = malloc(size);
ggml_backend_cann_transform(tensor, data, transform_buffer);
#ifndef NDEBUG
void *check_buffer = malloc(size);
ggml_backend_cann_transform_back(tensor, transform_buffer,
check_buffer);
GGML_ASSERT(memcmp(data, check_buffer, size) == 0);
free(check_buffer);
#endif
ACL_CHECK(aclrtMemcpy((char *)tensor->data + offset, size,
transform_buffer, size,
ACL_MEMCPY_HOST_TO_DEVICE));
@ -1517,13 +1510,6 @@ static void ggml_backend_cann_set_tensor_async(ggml_backend_t backend,
void *transform_buffer = malloc(size);
ggml_backend_cann_transform(tensor, data, transform_buffer);
#ifndef NDEBUG
void *check_buffer = malloc(size);
ggml_backend_cann_transform_back(tensor, transform_buffer,
check_buffer);
GGML_ASSERT(memcmp(data, check_buffer, size));
free(check_buffer);
#endif
ACL_CHECK(aclrtMemcpyAsync(
(char *)tensor->data + offset, size, transform_buffer, size,
ACL_MEMCPY_HOST_TO_DEVICE, cann_ctx->stream()));