[BugFix] rollback max_tokens and min_tokens when continue to infer (#5052)

Co-authored-by: liqinrui <[email protected]>

Author: LiqinruiG

fastdeploy/entrypoints/engine_client.py

@@ -210,14 +210,7 @@ async def add_requests(self, task):
             task["prompt_token_ids_len"] = len(task["prompt_token_ids"])
             input_ids_len = task["prompt_token_ids_len"]
 
-            completion_token_len = len(task["completion_token_ids"]) if task.get("completion_token_ids") else 0
-            task["max_tokens"] = min(
-                self.max_model_len - input_ids_len, max(0, task.get("max_tokens") - completion_token_len)
-            )
-
-            if task.get("min_tokens") is not None:
-                task["min_tokens"] = max(1, task["min_tokens"] - completion_token_len)
-
+            task["max_tokens"] = min(self.max_model_len - input_ids_len, task.get("max_tokens"))
             min_tokens = task.get("min_tokens", 1)
             if "messages" in task:
                 del task["messages"]

fastdeploy/input/ernie4_5_vl_processor/ernie4_5_vl_processor.py

@@ -252,9 +252,7 @@ def process_request_dict(self, request, max_model_len=None):
         else:
             raise ValueError(f"Request must contain 'prompt', or 'messages': {request}")
 
-        completion_token_len = 0
         if request.get("completion_token_ids"):
-            completion_token_len = len(request.get("completion_token_ids"))
             self.append_completion_tokens(outputs, request["completion_token_ids"])
 
         outputs = self.pack_outputs(outputs)
@@ -271,9 +269,7 @@ def process_request_dict(self, request, max_model_len=None):
             request["max_tokens"] = max(1, max_model_len - len(request["prompt_token_ids"]))
             tmp_max_tokens = request["max_tokens"]
         else:
-            tmp_max_tokens = min(
-                max_model_len - len(request["prompt_token_ids"]), max(0, request["max_tokens"] - completion_token_len)
-            )
+            request["max_tokens"] = min(max_model_len - len(request["prompt_token_ids"]), request["max_tokens"])
         if request.get("reasoning_max_tokens") is None:
             request["reasoning_max_tokens"] = max(int(tmp_max_tokens * 0.8), 1)
         data_processor_logger.info(f"Processed request {request}")