minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Scheduler] fix chunked prefill not always being full (#25741)

Browse Source
This commit is contained in:
Hanming Lu
2026-05-20 15:03:11 -07:00
committed by GitHub
parent 1a17d753f1
commit 371b6c9ea0
1 changed files with 16 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/managers/schedule_policy.py
+16 -2
View File
@@ -860,7 +860,14 @@ class PrefillAdder:
if swa_needed >= self.rem_swa_tokens: if swa_needed >= self.rem_swa_tokens:
return AddReqResult.NO_TOKEN return AddReqResult.NO_TOKEN
if real_input_tokens >= self.rem_input_tokens and len(self.can_run_list) != 0: if (
self.rem_chunk_tokens is None
and len(self.can_run_list) != 0
and real_input_tokens >= self.rem_input_tokens
):
# If without chunked prefill:
# - if the can_run_list is not empty, we satisfy the constraint of (max_prefill_tokens)
# - if the can_run_list is empty, always accept the first prefill request
return AddReqResult.OTHER return AddReqResult.OTHER
with self._lock_node(req.last_node): with self._lock_node(req.last_node):
@@ -888,7 +895,14 @@ class PrefillAdder:
input_tokens = self.ceil_paged_tokens(req.extend_input_len) input_tokens = self.ceil_paged_tokens(req.extend_input_len)
if input_tokens >= self.rem_input_tokens and len(self.can_run_list) != 0: if (
self.rem_chunk_tokens is None
and len(self.can_run_list) != 0
and input_tokens >= self.rem_input_tokens
):
# If without chunked prefill:
# - if the can_run_list is not empty, we satisfy the constraint of (max_prefill_tokens)
# - if the can_run_list is empty, always accept the first prefill request
return AddReqResult.OTHER return AddReqResult.OTHER
if self.dllm_config is not None: if self.dllm_config is not None: