sdgoij/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-05-15 05:24:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #4 from gaugarg-nv/minor_fixes

Browse Source 
Fix the seg fault without NCCL
This commit is contained in:
Johannes Gäßler
2026-02-12 14:19:13 +01:00
committed by GitHub
parent 3fdd0b7a6e 9bb9d78368
commit b12a56351d
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -329,9 +329,9 @@ static ggml_cuda_device_info ggml_cuda_init() {
dev_ids[id] = id;
}
NCCL_CHECK(ncclCommInitAll(info.comms, info.device_count, dev_ids));
#endif // GGML_USE_NCCL
return info;
#endif // GGML_USE_NCCL
}
const ggml_cuda_device_info & ggml_cuda_info() {
@@ -1132,7 +1132,7 @@ bool ggml_backend_cuda_allreduce_tensor(ggml_backend_t * backends, struct ggml_t
#if !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA)
static bool warning_printed = false;
if (!warning_printed) {
GGML_LOG_WARN("%s: NVIDIA Collective Communications Library (NCCL) is unavailable, multi GPU performance will be suboptimal\n");
GGML_LOG_WARN("%s: NVIDIA Collective Communications Library (NCCL) is unavailable, multi GPU performance will be suboptimal\n", __func__);
warning_printed = true;
}
GGML_UNUSED_VARS(backends, tensors, n_backends);