[Bug fix] Fix perf in mixed deployment with yiyan adpater (#3703)

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

Author: rainyfly

fastdeploy/output/token_processor.py

@@ -360,6 +360,7 @@ def _process_sampling_with_logprob_batch_output(self):
                 metrics = RequestMetrics(
                     arrival_time=task.arrival_time,
                     inference_start_time=task.inference_start_time,
+                    model_execute_time=time.time() - task.inference_start_time,
                     first_token_time=time.time() - task.inference_start_time,
                     time_in_queue=task.schedule_start_time - task.preprocess_end_time,
                     preprocess_cost_time=task.preprocess_end_time - task.preprocess_start_time,
@@ -503,6 +504,7 @@ def _process_batch_output(self):
                 metrics = RequestMetrics(
                     arrival_time=task.arrival_time,
                     inference_start_time=task.inference_start_time,
+                    model_execute_time=time.time() - task.inference_start_time,
                     first_token_time=time.time() - task.inference_start_time,
                     time_in_queue=task.schedule_start_time - task.preprocess_end_time,
                     preprocess_cost_time=task.preprocess_end_time - task.preprocess_start_time,
@@ -514,6 +516,7 @@ def _process_batch_output(self):
             else:
                 metrics = RequestMetrics(
                     arrival_time=time.time(),
+                    model_execute_time=time.time() - task.inference_start_time,
                     request_start_time=task.arrival_time,
                 )
             self.number_of_output_tokens += len(token_ids)

fastdeploy/scheduler/local_scheduler.py

@@ -209,7 +209,7 @@ def calc_required_blocks(self, token_num, block_size):
         return (token_num + block_size - 1) // block_size
 
     def get_unhandled_request_num(self):
-        return len(self.requests)
+        return len(self.ids) - self.ids_read_cursor
 
     def get_requests(
         self,