[Perf] Minimax M3 - Support cross-layer allreduce-norm fusion (#47631)

Signed-off-by: wzhao18 <wzhao18.sz@gmail.com>
Co-authored-by: Yongye Zhu <zyy1102000@gmail.com>
This commit is contained in:
Wei Zhao
2026-07-07 21:16:32 -07:00
committed by GitHub
co-authored by Yongye Zhu
parent 80eb01e93d
commit 2afa3f7e95
7 changed files with 65 additions and 17 deletions
@@ -279,6 +279,7 @@ class DeepseekV2MoE(nn.Module):
config: DeepseekV2Config | DeepseekV3Config,
parallel_config: ParallelConfig,
quant_config: QuantizationConfig | None = None,
reduce_results: bool = True,
prefix: str = "",
apply_routed_scale_to_output: bool = False,
):
@@ -379,6 +380,7 @@ class DeepseekV2MoE(nn.Module):
enable_eplb=self.enable_eplb,
num_redundant_experts=self.n_redundant_experts,
is_sequence_parallel=self.is_sequence_parallel,
reduce_results=reduce_results,
n_shared_experts=config.n_shared_experts
if self.is_fusion_moe_shared_experts_enabled
else None,