修复CompressedTensors的w8a16的支持

vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py

@@ -730,7 +730,7 @@ class CompressedTensorsLinearMethod(LinearMethodBase):
         )
         if weights_scheme is not None:
             num_bits = weights_scheme.num_bits
-            if num_bits == 4:
+            if isinstance(layer.scheme, CompressedTensorsWNA16):
                 return layer.scheme.process_weights_after_loading(layer)   
 
         n=layer.weight.shape[0]