Merge branch 'v0.9.2-dev-fix' into 'v0.9.2-dev'

fix: 修复重复判断逻辑

See merge request dcutoolkit/deeplearing/vllm!406

vllm/v1/core/sched/scheduler.py

@@ -281,22 +281,21 @@ class Scheduler(SchedulerInterface):
                     num_draft_tokens=num_draft_tokens,
                     num_lookahead_tokens=self.num_lookahead_tokens)
                 if new_blocks is None:
-                    if new_blocks is None:
-                        if self.use_pp:
-                            preemptable_reqs = [r for r in self.running if
-                                                r.num_tokens_with_spec != r.num_computed_tokens]
-                        else:
-                            preemptable_reqs = self.running
-                        # The request cannot be scheduled.
-                        # Preempt the lowest-priority request.
-                        if self.policy == SchedulingPolicy.PRIORITY:
-                            preempted_req = max(
-                                preemptable_reqs,
-                                key=lambda r: (r.priority, r.arrival_time),
-                            )
-                        else:
-                            preempted_req = preemptable_reqs[-1]
-                        self.running.remove(preempted_req)
+                    if self.use_pp:
+                        preemptable_reqs = [r for r in self.running if
+                                            r.num_tokens_with_spec != r.num_computed_tokens]
+                    else:
+                        preemptable_reqs = self.running
+                    # The request cannot be scheduled.
+                    # Preempt the lowest-priority request.
+                    if self.policy == SchedulingPolicy.PRIORITY:
+                        preempted_req = max(
+                            preemptable_reqs,
+                            key=lambda r: (r.priority, r.arrival_time),
+                        )
+                    else:
+                        preempted_req = preemptable_reqs[-1]
+                    self.running.remove(preempted_req)
 
                     self.kv_cache_manager.free(preempted_req)
                     preempted_req.status = RequestStatus.PREEMPTED