[Distributed] Enable FlashInfer all-reduce by default (#52998)

Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
Woosuk Kwon
2026-08-19 18:17:43 -07:00
committed by GitHub
co-authored by OpenAI Codex
parent 58e5ee0158
commit 6a962071bd
2 changed files with 6 additions and 3 deletions
@@ -58,7 +58,10 @@ class CudaCommunicator(DeviceCommunicatorBase):
use_custom_allreduce = _ENABLE_CUSTOM_ALL_REDUCE
use_torch_symm_mem = envs.VLLM_ALLREDUCE_USE_SYMM_MEM
use_flashinfer_allreduce = envs.VLLM_ALLREDUCE_USE_FLASHINFER
# FlashInfer all-reduce does not provide a fixed reduction order.
use_flashinfer_allreduce = (
envs.VLLM_ALLREDUCE_USE_FLASHINFER and not envs.VLLM_BATCH_INVARIANT
)
use_aiter_allreduce = use_custom_allreduce and bool(
rocm_aiter_ops.is_custom_all_reduce_enabled()
)