CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-15 11:17:31 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: fix crash with partial offloading of MoE (#13439)

Browse Source
This commit is contained in:
Johannes Gäßler
2025-05-11 16:09:33 +02:00
committed by GitHub
parent 7f323a589f
commit 7474e00b34
3 changed files with 12 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-cuda/mmvq.cu
View File

@@ -515,11 +515,11 @@ void ggml_cuda_mul_mat_vec_q(
// If src0 is a temporary compute buffer, clear any potential padding.
if (ggml_backend_buffer_get_usage(src0->buffer) == GGML_BACKEND_BUFFER_USAGE_COMPUTE) {
GGML_ASSERT(ggml_is_contiguously_allocated(src0));
GGML_ASSERT(!src0->view_src);
const size_t size_data = ggml_nbytes(src0);
const size_t size_alloc = ggml_backend_buffer_get_alloc_size(src0->buffer, src0);
if (size_alloc > size_data) {
GGML_ASSERT(ggml_is_contiguously_allocated(src0));
GGML_ASSERT(!src0->view_src);
CUDA_CHECK(cudaMemsetAsync((char *) src0->data + size_data, 0, size_alloc - size_data, stream));
}
}