slaren commited on
Commit
b7f6691
·
1 Parent(s): 3519475

cuda : clear error after buffer allocation failure (llama/7376)

Browse files
Files changed (1) hide show
  1. ggml-cuda.cu +2 -0
ggml-cuda.cu CHANGED
@@ -539,6 +539,8 @@ GGML_CALL static ggml_backend_buffer_t ggml_backend_cuda_buffer_type_alloc_buffe
539
  void * dev_ptr;
540
  cudaError_t err = cudaMalloc(&dev_ptr, size);
541
  if (err != cudaSuccess) {
 
 
542
  GGML_CUDA_LOG_ERROR("%s: allocating %.2f MiB on device %d: cudaMalloc failed: %s\n", __func__, size / 1024.0 / 1024.0, buft_ctx->device, cudaGetErrorString(err));
543
  return nullptr;
544
  }
 
539
  void * dev_ptr;
540
  cudaError_t err = cudaMalloc(&dev_ptr, size);
541
  if (err != cudaSuccess) {
542
+ // clear the error
543
+ cudaGetLastError();
544
  GGML_CUDA_LOG_ERROR("%s: allocating %.2f MiB on device %d: cudaMalloc failed: %s\n", __func__, size / 1024.0 / 1024.0, buft_ctx->device, cudaGetErrorString(err));
545
  return nullptr;
546
  }