[BugFix] Modify the bug in Qwen2 when enabling ENABLE_V1_KVCACHE_SCHEDULER. (#3670)

* merge 2.1

* fix

* pre-commit

* fix

fastdeploy/engine/engine.py

@@ -29,6 +29,7 @@ import traceback
 import uuid
 import weakref
 from concurrent.futures import ThreadPoolExecutor
+from dataclasses import asdict
 from typing import Dict, List, Optional, Tuple
 
 import numpy as np
@@ -479,6 +480,8 @@ class LLMEngine:
         """
         # TODO 输入输出长度确认
 
+        if sampling_params is not None:
+            task.update(asdict(sampling_params))
         request = Request.from_dict(task)
         llm_logger.info(f"Receive request {request}")
         if sampling_params is not None:

fastdeploy/worker/gpu_model_runner.py

@@ -260,7 +260,11 @@ class GPUModelRunner(ModelRunnerBase):
                     self.share_inputs["need_think_end"][idx : idx + 1, :] = 0
                     self.share_inputs["reasoning_index"][idx : idx + 1, :] = 0
 
+                if len(request.output_token_ids) == 0:
+                    input_ids = request.prompt_token_ids
+                else:
                     input_ids = request.prompt_token_ids + request.output_token_ids
+
                 logger.debug(
                     f"Handle prefill request {request} at idx {idx} prefill_start_index {prefill_start_index} prefill_end_index {prefill_end_index} need_prefilled_token_num {len(input_ids)}"
                 )