fix bugs in attention.py and request_handler.py

colossalai/inference/core/engine.py

@@ -214,9 +214,6 @@ class InferenceEngine:
             List[str]: Decoded finished sequences generated by one step.
         """
 
-        if self.verbose:
-            self.logger.info("Running generation step")
-
         output_list = []
         batch = self.request_handler.schedule()
 
@@ -224,6 +221,7 @@ class InferenceEngine:
             batch,
             self.k_cahce,
             self.v_cache,
+            padding_id=self.tokenizer.pad_token_id,
         )
 
         logits = logits[:, -1, :]