[ROCm] [Fused Moe EP] Use binary expert mask for aiter fused moe kernel (#29773)

Signed-off-by: ZhiweiYan-96 <zhiwei.yan@amd.com>
This commit is contained in:
Zhiwei
2025-12-08 05:23:46 +00:00
committed by GitHub
parent d726a7b0ed
commit c6df05ebb4
2 changed files with 5 additions and 0 deletions
@@ -633,6 +633,7 @@ class QuarkOCP_MX_MoEMethod(QuarkMoEMethod):
topk_ids=topk_ids,
activation=activation,
quant_config=self.moe_quant_config,
expert_map=expert_map,
)
else:
from vllm.model_executor.layers.fused_moe import fused_experts