use marlin remove:

rshaw@neuralmagic.com · rshaw@neuralmagic.com · commit 183bfe77b499 · 2024-07-20T16:58:43.000Z
diff --git a/vllm/model_executor/layers/quantization/fbgemm_fp8.py b/vllm/model_executor/layers/quantization/fbgemm_fp8.py
@@ -40,8 +40,6 @@ def __init__(self, ignore_list: List[str], input_scale_ub: float):
         capability = capability[0] * 10 + capability[1]
         self.use_marlin = capability < 89
 
-        self.use_marlin = True
-
     @classmethod
     def get_name(cls) -> str:
         return "fbgemm_fp8"
