[Bugfix] Restore portable all2all backend default (#53952)

Signed-off-by: Kevin Luu <51931015+khluu@users.noreply.github.com>
Co-authored-by: Kevin Luu <51931015+khluu@users.noreply.github.com>
This commit is contained in:
vllm-agent
2026-08-26 16:19:49 -07:00
committed by GitHub
co-authored by Kevin Luu
parent 3fd72ffd7b
commit d1e5e66ee3
2 changed files with 5 additions and 1 deletions
+1 -1
View File
@@ -192,7 +192,7 @@ class ParallelConfig:
with 4 experts and 2 ranks, rank 0 will have experts [0, 2] and rank 1
will have experts [1, 3]. This strategy can help improve load balancing
for grouped expert models with no redundant experts."""
all2all_backend: All2AllBackend = "flashinfer_nvlink_one_sided"
all2all_backend: All2AllBackend = "allgather_reducescatter"
"""All2All backend for MoE expert parallel communication. Available options:
- "allgather_reducescatter": All2all based on allgather and reducescatter