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

prevent server crash from potential invalid grammar (#7897)

Browse Source
This commit is contained in:
ehuaa
2025-07-15 11:21:45 +08:00
committed by GitHub
parent 7c39e8a198
commit 64e78bb31b
1 changed files with 21 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
python/sglang/srt/managers/scheduler_output_processor_mixin.py
View File

@@ -7,7 +7,7 @@ from typing import TYPE_CHECKING, List, Optional, Tuple, Union
from sglang.srt.disaggregation.utils import DisaggregationMode
from sglang.srt.layers.logits_processor import LogitsProcessorOutput
from sglang.srt.managers.io_struct import BatchEmbeddingOut, BatchTokenIDOut
from sglang.srt.managers.io_struct import AbortReq, BatchEmbeddingOut, BatchTokenIDOut
from sglang.srt.managers.schedule_batch import BaseFinishReason, Req, ScheduleBatch
if TYPE_CHECKING:
@@ -126,7 +126,16 @@ class SchedulerOutputProcessorMixin:
)
if req.grammar is not None:
req.grammar.accept_token(next_token_id)
# FIXME: this try-except block is for handling unexpected xgrammar issue.
try:
req.grammar.accept_token(next_token_id)
except ValueError as e:
# Grammar accept_token can raise ValueError if the token is not in the grammar.
# This can happen if the grammar is not set correctly or the token is invalid.
logger.error(
f"Grammar accept_token failed for req {req.rid} with token {next_token_id}: {e}"
)
self.abort_request(AbortReq(req.rid))
req.grammar.finished = req.finished()
else:
# being chunked reqs' prefill is not finished
@@ -263,7 +272,16 @@ class SchedulerOutputProcessorMixin:
)
if req.grammar is not None and batch.spec_algorithm.is_none():
req.grammar.accept_token(next_token_id)
# FIXME: this try-except block is for handling unexpected xgrammar issue.
try:
req.grammar.accept_token(next_token_id)
except ValueError as e:
# Grammar accept_token can raise ValueError if the token is not in the grammar.
# This can happen if the grammar is not set correctly or the token is invalid.
logger.error(
f"Grammar accept_token failed for req {req.rid} with token {next_token_id}: {e}"
)
self.abort_request(AbortReq(req.rid))
req.grammar.finished = req.finished()
self.set_next_batch_sampling_info_done(batch)