[ Misc ] fbgemm checkpoints (#6559)

2024-07-20 12:36:57 -04:00
parent 9042d68362
commit 683e3cb9c4
24 changed files with 234 additions and 47 deletions
--- a/vllm/model_executor/layers/fused_moe/layer.py
+++ b/vllm/model_executor/layers/fused_moe/layer.py
@@ -182,7 +182,7 @@ class FusedMoE(torch.nn.Module):
            self.quant_method: Optional[QuantizeMethodBase] = (
                UnquantizedFusedMoEMethod())
        else:
-            self.quant_method = quant_config.get_quant_method(self)
+            self.quant_method = quant_config.get_quant_method(self, prefix)
        assert self.quant_method is not None

        self.quant_method.create_weights(