[Misc] Fix incorrect layer type annotation in Fp8LinearMethod (#46818)
Signed-off-by: shaojinjie.sjj <shaojinjiesjj@gmail.com> Co-authored-by: shaojinjie.sjj <shaojinjiesjj@gmail.com>
jj shao committed
Jun 28, 2026 at 04:23 UTC
ea2ead1db33dafb067aa64d4ce7b9c2150c12091
1 file changed
+2
-2
vllm/model_executor/layers/quantization/fp8.py
+2
-2
@@ -320,7 +320,7 @@ class Fp8LinearMethod(LinearMethodBase):
320
321
def create_weights(
322
self,
323
- layer: RoutedExperts,
323
+ layer: torch.nn.Module,
324
input_size_per_partition: int,
325
output_partition_sizes: list[int],
326
input_size: int,
@@ -394,7 +394,7 @@ class Fp8LinearMethod(LinearMethodBase):
394
395
self.use_marlin = isinstance(self.fp8_linear, MarlinFP8ScaledMMLinearKernel)
396
397
- def process_weights_after_loading(self, layer: RoutedExperts) -> None:
397
+ def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
398
if self.use_marlin:
399
if not self.block_quant:
400
# Canonicalize to (K, N) for the kernel.