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

Fix/partial gen from waiting queue miss metadata (#17610)

Browse Source
This commit is contained in:
JD
2026-02-13 19:04:08 -08:00
committed by GitHub
parent 45a4697d45
commit f6c18c3a85
2 changed files with 40 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
python/sglang/srt/managers/tokenizer_manager.py
View File

@@ -2164,6 +2164,7 @@ class TokenizerManager(TokenizerCommunicatorMixin, TokenizerManagerMultiItemMixi
return
state = self.rid_to_state[recv_obj.rid]
state.finished = True
state.finished_time = time.time()
abort_message = recv_obj.abort_message or "Abort in waiting queue"
finish_reason = {
@@ -2172,7 +2173,12 @@ class TokenizerManager(TokenizerCommunicatorMixin, TokenizerManagerMultiItemMixi
}
if recv_obj.finished_reason:
finish_reason = recv_obj.finished_reason
meta_info = {"id": recv_obj.rid, "finish_reason": finish_reason}
meta_info = {
"id": recv_obj.rid,
"finish_reason": finish_reason,
"weight_version": self.server_args.weight_version,
"e2e_latency": state.finished_time - state.created_time,
}
is_stream = getattr(state.obj, "stream", False)
if getattr(state.obj, "return_logprob", False):
self.add_logprob_to_meta_info(