update blockwise_int8.py

vllm/model_executor/layers/quantization/blockwise_int8.py

@@ -139,7 +139,6 @@ class BlockInt8LinearMethod(LinearMethodBase):
         
         assert self.quant_config.weight_block_size is not None
         assert self.quant_config.is_checkpoint_int8_serialized
-        self.tritonsingleton= W8a8GetCacheJSON()
 
     def create_weights(
         self,
@@ -337,6 +336,7 @@ class BlockInt8MoEMethod:
         self.quant_config = quant_config
         assert self.quant_config.weight_block_size is not None
         assert self.quant_config.is_checkpoint_int8_serialized
+        self.tritonsingleton= W8a8GetCacheJSON()
 
     def create_weights(
         self,