[Scheduler] Unify per-iteration request intake into ingest_requests() (#38389)

This commit is contained in:
Liangsheng Yin
2026-09-07 20:02:02 -07:00
committed by GitHub
parent 2bf04f3a67
commit b23d835048
12 changed files with 44 additions and 56 deletions
@@ -114,8 +114,7 @@ class SchedulerMultiplexMixin:
while True:
with torch.cuda.stream(decode_stream):
set_pdmux_status(False)
recv_reqs = self.request_receiver.recv_requests()
self.process_input_requests(recv_reqs)
self.ingest_requests()
running_batch = self.running_batch
with torch.cuda.stream(prefill_stream):