[Bugfix] Fix weight loading issues caused by #41184 (#45054)

Signed-off-by: Bill Nell <bnell@redhat.com>
This commit is contained in:
bnellnm
2026-06-10 01:20:13 -04:00
committed by GitHub
parent 2c9c07c85e
commit f4966f8b3d
5 changed files with 55 additions and 25 deletions
+2 -6
View File
@@ -8,7 +8,6 @@ from typing import TYPE_CHECKING
from vllm.config import get_current_vllm_config
from vllm.model_executor.layers.fused_moe import (
MoERunner,
RoutedExperts,
UnquantizedFusedMoEMethod,
)
@@ -133,7 +132,7 @@ class DeepseekV4FP8Config(Fp8Config):
return None
def get_quant_method(self, layer, prefix):
if isinstance(layer, (MoERunner, RoutedExperts)):
if isinstance(layer, RoutedExperts):
if is_layer_skipped(
prefix=prefix,
ignored_layers=self.ignored_layers,
@@ -156,9 +155,6 @@ class DeepseekV4FP8Config(Fp8Config):
return super().get_quant_method(layer, prefix)
def is_mxfp4_quant(self, prefix, layer):
if (
not isinstance(layer, (MoERunner, RoutedExperts))
or self.expert_dtype != "fp4"
):
if not isinstance(layer, RoutedExperts) or self.expert_dtype != "fp4":
return False
return self.moe_quant_algo != "NVFP4"