From ea2ead1db33dafb067aa64d4ce7b9c2150c12091 Mon Sep 17 00:00:00 2001 From: jj shao Date: Sun, 28 Jun 2026 04:23:59 +0800 Subject: [PATCH] [Misc] Fix incorrect layer type annotation in Fp8LinearMethod (#46818) Signed-off-by: shaojinjie.sjj Co-authored-by: shaojinjie.sjj --- vllm/model_executor/layers/quantization/fp8.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/vllm/model_executor/layers/quantization/fp8.py b/vllm/model_executor/layers/quantization/fp8.py index 869fbf7523..d4ec1c093a 100644 --- a/vllm/model_executor/layers/quantization/fp8.py +++ b/vllm/model_executor/layers/quantization/fp8.py @@ -320,7 +320,7 @@ class Fp8LinearMethod(LinearMethodBase): def create_weights( self, - layer: RoutedExperts, + layer: torch.nn.Module, input_size_per_partition: int, output_partition_sizes: list[int], input_size: int, @@ -394,7 +394,7 @@ class Fp8LinearMethod(LinearMethodBase): self.use_marlin = isinstance(self.fp8_linear, MarlinFP8ScaledMMLinearKernel) - def process_weights_after_loading(self, layer: RoutedExperts) -> None: + def process_weights_after_loading(self, layer: torch.nn.Module) -> None: if self.use_marlin: if not self.block_quant: # Canonicalize to (K, N) for the kernel.