]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
CUDA: manage NCCL communicators in context (#21891)
authorJohannes Gäßler <redacted>
Wed, 15 Apr 2026 13:58:40 +0000 (15:58 +0200)
committerGitHub <redacted>
Wed, 15 Apr 2026 13:58:40 +0000 (15:58 +0200)
commit014dca49d6c1c735d58f8bcf4e101f8cc80fbfc5
tree44715dee2ec40fe0c380f35b5ff93ef19f465e55
parentadb541a6ad077d037edcdca346c6c9624b2aac66
CUDA: manage NCCL communicators in context (#21891)

* CUDA: manage NCCL communicators in context

* add check that all backends are CUDA

* remove unused vector, limit init to > 1 GPUs

* fix warnings

* fix cuda device, cache allreduce
ggml/include/ggml-backend.h
ggml/src/ggml-backend-meta.cpp
ggml/src/ggml-cuda/common.cuh
ggml/src/ggml-cuda/ggml-cuda.cu