From 4d4ae1c9a15296adc5edb9593ac4edbb7e7a4223 Mon Sep 17 00:00:00 2001 From: leo-pony Date: Mon, 21 Oct 2024 19:13:21 +0800 Subject: [PATCH] Fix the Bug: inference running result is garbled in debug running model for LM models who's type is Q4_0 class --- ggml/src/ggml-cann.cpp | 14 -------------- 1 file changed, 14 deletions(-) diff --git a/ggml/src/ggml-cann.cpp b/ggml/src/ggml-cann.cpp index 6365d0d66..3ebea6702 100644 --- a/ggml/src/ggml-cann.cpp +++ b/ggml/src/ggml-cann.cpp @@ -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()));