EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Ensure chunked request extension length respects both rem_chunk_tokens and rem_total_tokens limits (#10003)

Browse Source
This commit is contained in:
pansicheng
2025-09-04 19:15:26 +08:00
committed by GitHub
parent 27e8ffed37
commit c67569491c
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
python/sglang/srt/managers/schedule_policy.py
View File

@@ -380,8 +380,9 @@ class PrefillAdder:
self.log_input_tokens += extend_input_len self.log_input_tokens += extend_input_len
def add_chunked_req(self, req: Req): def add_chunked_req(self, req: Req):
truncated = req.extend_input_len > self.rem_chunk_tokens _rem_tokens = min(self.rem_chunk_tokens, int(self.rem_total_tokens))
req.extend_input_len = min(req.extend_input_len, self.rem_chunk_tokens) truncated = req.extend_input_len > _rem_tokens
req.extend_input_len = min(req.extend_input_len, _rem_tokens)
req.fill_ids = req.fill_ids[: len(req.prefix_indices) + req.extend_input_len] req.fill_ids = req.fill_ids[: len(req.prefix_indices) + req.extend_input_len]
self.can_run_list.append(req) self.can_run_list.append(req)
self._update_prefill_budget( self._update_prefill_budget(