A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Bugfix][Core]Fix block table out-of-range issue in priority scheduling (#26661)
Signed-off-by: quanliu <18646313696@163.com>
quanliu committed
Oct 13, 2025 at 09:25 UTC
41f3884438c082c4cc2250eb800b8586e1a103d8
1 file changed
+3
vllm/v1/core/sched/scheduler.py
+3
@@ -271,6 +271,9 @@ class Scheduler(SchedulerInterface):
271
self.running.remove(preempted_req)
272
if preempted_req in scheduled_running_reqs:
273
scheduled_running_reqs.remove(preempted_req)
274
+ token_budget += num_scheduled_tokens[preempted_req.request_id]
275
+ req_to_new_blocks.pop(preempted_req.request_id)
276
+ num_scheduled_tokens.pop(preempted_req.request_id)
277
else:
278
preempted_req = self.running.pop()
279