Public/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-19 01:04:55 +02:00
Code Issues Packages Projects Releases Wiki Activity

allreduce: assert n_backends == 2 instead of soft-fallback

Browse Source 
ar_pipeline_init already requires n_devices == 2 and bails before any AR can
get here, so by the time we reach try_allreduce_internal we know we have
exactly two backends.  Replace the runtime-debug-log fallback with a hard
assert.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
 NCCL is in use.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Scott Cutler
2026-05-04 20:15:22 -07:00
parent a19d711385
commit 5d1b22ab3d
1 changed files with 1 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/ggml-cuda.cu
+1 -4
View File
@@ -1287,10 +1287,7 @@ static bool ggml_backend_cuda_comm_try_allreduce_internal(
const int64_t ne = ggml_nelements(tensors[0]);
const ggml_type type = tensors[0]->type;
if (n_backends != 2) {
GGML_LOG_DEBUG("%s: internal unsupported: n_backends=%zu\n", __func__, n_backends);
return false;
}
GGML_ASSERT(n_backends == 2);
if (type != GGML_TYPE_F32 && type != GGML_TYPE_F16 && type != GGML_TYPE_BF16) {
GGML_LOG_DEBUG("%s: internal unsupported: type=%d\n", __func__, (int) type);
return false;