diff options
author | Henrik Forstén <henrik.forsten@gmail.com> | 2023-12-22 15:34:05 +0200 |
---|---|---|
committer | GitHub <noreply@github.com> | 2023-12-22 14:34:05 +0100 |
commit | 6724ef16573ec7ecce620be56cbbff145856b2fb (patch) | |
tree | e3746c4b940a6c1c343e5d93259ab6cb548df7af | |
parent | 48b7ff193e64c97ab174280ba0eb8d14b47c49ba (diff) |
Fix CudaMemcpy direction (#4599)
-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 37d7f279..da8fd1e0 100644 --- a/ggml-cuda.cu +++ b/ggml-cuda.cu @@ -8843,7 +8843,7 @@ static void ggml_cuda_mul_mat_id(const ggml_tensor * src0, const ggml_tensor * s const cudaMemcpyKind src1_kind = src1->backend == GGML_BACKEND_CPU ? cudaMemcpyHostToDevice : cudaMemcpyDeviceToDevice; const cudaMemcpyKind dst_kind = dst->backend == GGML_BACKEND_CPU ? - cudaMemcpyHostToDevice : cudaMemcpyDeviceToDevice; + cudaMemcpyDeviceToHost : cudaMemcpyDeviceToDevice; for (int32_t row_id = 0; row_id < n_as; ++row_id) { const struct ggml_tensor * src0_row = dst->src[row_id + 2]; |