[Bugfix] Fix FP8 Marlin MoE and enable for compressed-tensors models (#18026)

Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2025-05-13 19:48:33 -07:00
committed by GitHub
parent 6266c57bae
commit 9a2a6357de
3 changed files with 53 additions and 6 deletions
@@ -811,6 +811,10 @@ class Fp8MoEMethod(FusedMoEMethodBase):
)
if self.use_marlin:
assert activation == "silu", (
f"{activation} not supported for Marlin MoE.")
assert not apply_router_weight_on_input, (
"Apply router weight on input not supported for Marlin MoE.")
return torch.ops.vllm.fused_marlin_moe(
x,
layer.w13_weight,