[MoE Refactor] Migrate ModelOptMxFp8FusedMoE to oracle (#42768)

Signed-off-by: Bill Nell <bnell@redhat.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
This commit is contained in:
bnellnm
2026-05-26 11:14:14 -04:00
committed by GitHub
co-authored by Robert Shaw
parent 6ab6ffb428
commit b226ddacfd
8 changed files with 114 additions and 130 deletions
@@ -105,9 +105,9 @@ class Int8OnlineMoEMethod(OnlineMoEMethodBase):
def get_fused_moe_quant_config(
self, layer: torch.nn.Module
) -> "FusedMoEQuantConfig | None":
quant_config = make_int8_moe_quant_config(
return make_int8_moe_quant_config(
w1_scale=layer.w13_scale,
w2_scale=layer.w2_scale,
w1_bias=getattr(layer, "w13_bias", None),
w2_bias=getattr(layer, "w2_bias", None),
)
self._maybe_inject_biases(quant_config, layer)
return quant_config