[Model] Apply shared experts overlap optimization to all models with shared experts (#26145)

Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2025-10-09 11:31:04 -04:00
committed by GitHub
parent 3b736e1c38
commit 47e66c24e2
15 changed files with 285 additions and 297 deletions
@@ -741,6 +741,8 @@ class Fp8MoEMethod(FusedMoEMethodBase):
layer.w13_input_scale = None
layer.w2_input_scale = None
self.rocm_aiter_moe_enabled = False
def process_weights_after_loading(self, layer: Module) -> None:
# Lazy import to avoid importing triton too early.
from vllm.model_executor.layers.fused_moe.rocm_aiter_fused_moe import (