kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-02-04 18:21:52 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

[None][fix] Always reset drafting states for GuidedDecoder (#10899)

Browse Source 
Signed-off-by: Enwei Zhu <21126786+syuoni@users.noreply.github.com>
Signed-off-by: Wangshanshan <30051912+dominicshanshan@users.noreply.github.com>
This commit is contained in:
Enwei Zhu 2026-01-22 21:15:19 +08:00 committed by Yanchao Lu
parent fafc22e3d4
commit ccdd8461ac
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
tensorrt_llm/_torch/pyexecutor/guided_decoder.py
View File

@ -276,6 +276,7 @@ class GuidedDecoder:
assert len(req.draft_tokens) == 0
self.num_advanced_draft_tokens[
slot] += self.num_advanced_tokens[slot]
except Exception as e:
error_msg = f"Guided decoding error: {str(e)}"
failed_requests.append((req.request_id, error_msg))
@ -406,10 +407,9 @@ class GuidedDecoder:
for req in requests.valid_requests():
slot = req.seq_slot
if self.num_advanced_draft_tokens[slot] <= 0:
continue
self.grammar_matchers[slot].rollback(
self.num_advanced_draft_tokens[slot])
if self.num_advanced_draft_tokens[slot] > 0:
self.grammar_matchers[slot].rollback(
self.num_advanced_draft_tokens[slot])
# Reset the drafting states.
self.num_advanced_draft_tokens[slot] = 0
self.is_draft_terminated[slot] = False