linfeng-yuan 7c9aa498d6 [releases/v0.18.0][BugFix] Restore global_bs=0 and mc2_mask for uniform-token dispatching and support inter-node roce hierarchical MC2 communication (#8040) ...

### What this PR does / why we need it? 
Cherry-picked from #8039 
Restore the setting of MC2 `global_bs` and `mc2_mask` handling when
`all_reduce` across DP group cannot be skipped. Ascend MC2 ops require
`global_bs=0` + `mc2_mask` while enabling inter-node roce hierarchical
communication. PR #4983 always passed non-zero `global_bs` without
`mc2_mask`, which is incompatible with hierarchy comm raised in PR #7583
  **Changes:**                           
- Add `should_skip_allreduce_across_dp_group()` to `utils.py` with
hierarchy constraint
- Set `global_bs=0` when allreduce is not skipped; pass `mc2_mask`
accordingly
- Add `mc2_mask` field to `MoEMC2CombineMetadata` for dispatch→combine
propagation
### Does this PR introduce _any_ user-facing change?
No. But this PR fixes cross-super-node communication function on A3 with
`enable_mc2_hierarchy_comm=True` in `additional_config` and `export
HCCL_INTRA_ROCE_ENABLE=1`.

### How was this patch tested?
E2E serving succeeded and CI pssed.

- vLLM version: v0.18.0
- vLLM main:
14acf429ac

---------

Signed-off-by: linfeng-yuan <1102311262@qq.com>

2026-04-09 16:51:17 +08:00

..

__init__.py

[Refactor] [MoE] Rename moe-related classes & files (#3646)

2025-10-25 11:22:03 +08:00

comm_utils.py

[Lint]Style: Convert vllm-ascend/ to ruff format(Batch #11) (#6176)

2026-02-06 15:28:49 +08:00

experts_selector.py

[A5][bugfix] Fix fused MoE A5 MXFP8 scale normalization, load-balance routing and gating_topk ops (#7573)

2026-03-25 17:20:28 +08:00

fused_moe.py

Main2main Upgrade vllm commit to 0320 17:00 (#7510)

2026-03-23 21:37:41 +08:00

moe_comm_method.py

[refactor] replace scattered business kwargs with typed request objects and explicit stage boundaries (#7024)

2026-03-20 23:23:57 +08:00

moe_mlp.py

[A5][bugfix] Fix fused MoE A5 MXFP8 scale normalization, load-balance routing and gating_topk ops (#7573)

2026-03-25 17:20:28 +08:00

moe_runtime_args.py

[refactor] replace scattered business kwargs with typed request objects and explicit stage boundaries (#7024)

2026-03-20 23:23:57 +08:00

moe_stage_contracts.py

[releases/v0.18.0][BugFix] Restore global_bs=0 and mc2_mask for uniform-token dispatching and support inter-node roce hierarchical MC2 communication (#8040)

2026-04-09 16:51:17 +08:00

moe_stage_params.py

[refactor] replace scattered business kwargs with typed request objects and explicit stage boundaries (#7024)

2026-03-20 23:23:57 +08:00

prepare_finalize.py

[Feat][SP] Suport SP for VL MoE models (#7044)

2026-03-24 17:16:00 +08:00

token_dispatcher.py

[releases/v0.18.0][BugFix] Restore global_bs=0 and mc2_mask for uniform-token dispatching and support inter-node roce hierarchical MC2 communication (#8040)

2026-04-09 16:51:17 +08:00