diff --git a/server/text_generation_server/layers/moe/gptq_marlin.py b/server/text_generation_server/layers/moe/gptq_marlin.py index 3d4ca9d8..d909f397 100644 --- a/server/text_generation_server/layers/moe/gptq_marlin.py +++ b/server/text_generation_server/layers/moe/gptq_marlin.py @@ -69,7 +69,11 @@ class GPTQMarlinSparseMoELayer(nn.Module): gate_proj_name: str = "gate_proj", up_proj_name: str = "up_proj", down_proj_name: str = "down_proj", + scoring_func: Optional[str] = None, + e_score_correction_bias: Optional[float] = None, ): + assert scoring_func is None, "scoring func is not handled" + assert e_score_correction_bias is None, "scoring correction bias is not handled" super().__init__() if not (