Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-26 03:12:23 +01:00
Code Issues Packages Projects Releases Wiki Activity

#1869 Fix null reference errors when training from scratch with CUDA (#1907)

Browse Source 
* #1869 Fix null reference errors when training from scratch with CUDA build

Calling ggml_compute_forward when node->src0 was null was causing train-text-from-scratch.exe to terminate unexpectedly.

* ggml : do not dereference src0 if NULL

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Robyn 2023-06-25 04:10:29 +10:00 committed by GitHub
parent 65bdd52a86
commit 5ec8dd5a3c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
ggml-cuda.cu
View File

@ -2635,7 +2635,7 @@ void ggml_cuda_free_scratch() {
bool ggml_cuda_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor){ bool ggml_cuda_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor){
ggml_cuda_func_t func; ggml_cuda_func_t func;
const bool any_on_device = tensor->backend == GGML_BACKEND_GPU const bool any_on_device = tensor->backend == GGML_BACKEND_GPU
|| tensor->src0->backend == GGML_BACKEND_GPU || tensor->src0->backend == GGML_BACKEND_GPU_SPLIT || (tensor->src0 != nullptr && (tensor->src0->backend == GGML_BACKEND_GPU || tensor->src0->backend == GGML_BACKEND_GPU_SPLIT))
|| (tensor->src1 != nullptr && tensor->src1->backend == GGML_BACKEND_GPU); || (tensor->src1 != nullptr && tensor->src1->backend == GGML_BACKEND_GPU);
switch (tensor->op) { switch (tensor->op) {

2
ggml.c
View File

@ -14911,7 +14911,7 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
if (skip_cpu) { if (skip_cpu) {
return; return;
} }
GGML_ASSERT(tensor->src0->backend == GGML_BACKEND_CPU); GGML_ASSERT(tensor->src0 == NULL || tensor->src0->backend == GGML_BACKEND_CPU);
GGML_ASSERT(tensor->src1 == NULL || tensor->src1->backend == GGML_BACKEND_CPU); GGML_ASSERT(tensor->src1 == NULL || tensor->src1->backend == GGML_BACKEND_CPU);
#endif // GGML_USE_CUBLAS #endif // GGML_USE_CUBLAS