Merge branch 'v0.11.0-dev-wm' into 'v0.11.0-dev'

[fix]修复splitpd调度lora报错

See merge request dcutoolkit/deeplearing/vllm!432

vllm/v1/core/sched/scheduler.py

@@ -663,6 +663,9 @@ class Scheduler(SchedulerInterface):
         # For logging.
         scheduled_timestamp = time.monotonic()
 
+        # Record the LoRAs in scheduled_running_reqs
+        scheduled_loras: set[int] = set()
+
         # Use a temporary RequestQueue to collect requests that need to be
         # skipped and put back at the head of the waiting queue later
         skipped_waiting_requests = create_request_queue(self.policy)
@@ -993,13 +996,11 @@ class Scheduler(SchedulerInterface):
                         self.encoder_cache_manager.allocate(request, i)
                     encoder_compute_budget = new_encoder_compute_budget
 
-            # Record the LoRAs in scheduled_running_reqs
-            scheduled_loras: set[int] = set()
-            if self.lora_config:
-                scheduled_loras = set(
-                    req.lora_request.lora_int_id for req in scheduled_running_reqs
-                    if req.lora_request and req.lora_request.lora_int_id > 0)
-                assert len(scheduled_loras) <= self.lora_config.max_loras
+        if self.lora_config:
+            scheduled_loras = set(
+                req.lora_request.lora_int_id for req in scheduled_running_reqs
+                if req.lora_request and req.lora_request.lora_int_id > 0)
+            assert len(scheduled_loras) <= self.lora_config.max_loras
 
         # Check if the scheduling constraints are satisfied.
         total_num_scheduled_tokens = sum(num_scheduled_tokens.values())
@@ -1089,17 +1090,14 @@ class Scheduler(SchedulerInterface):
     
     def schedule(self) -> SchedulerOutput:
         if envs.VLLM_USE_PD_SPLIT: 
+            return self.schedule_split_pd()
+        else:
             if self.connector is not None:
                 return self.schedule_default()
-            if self.use_mla:
-                if self.full_cuda_graph and self.num_spec_tokens > 0:
-                    return self.schedule_split_pd() 
-                else:
-                    return self.schedule_default()
-            else:
+            if self.full_cuda_graph and self.use_mla and self.num_spec_tokens > 0 :
                 return self.schedule_split_pd()
-        else:
-            return self.schedule_default()
+            else:
+                return self.schedule_default()
 
     def _update_after_schedule(
         self,