diff options
| author | Robyn <robyngraf@users.noreply.github.com> | 2023-06-25 04:10:29 +1000 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2023-06-24 20:10:29 +0200 |
| commit | 5ec8dd5a3c6a9a109351d2257bb9d53869bd0a94 (patch) | |
| tree | 571f9f195508ff20b7d23aa698c2660d0d0f8ab5 /ggml-cuda.cu | |
| parent | 65bdd52a867539691007f85c5508146d507f72c1 (diff) | |
#1869 Fix null reference errors when training from scratch with CUDA (#1907)
* #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>
Diffstat (limited to 'ggml-cuda.cu')
| -rw-r--r-- | ggml-cuda.cu | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/ggml-cuda.cu b/ggml-cuda.cu index 36a251e..010682e 100644 --- a/ggml-cuda.cu +++ b/ggml-cuda.cu @@ -2635,7 +2635,7 @@ void ggml_cuda_free_scratch() { bool ggml_cuda_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor){ ggml_cuda_func_t func; 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); switch (tensor->op) { |