A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Bugfix] Catch and log invalid token ids in detokenizer #2 (#26445)
Signed-off-by: Nick Hill <nhill@redhat.com>
Nick Hill committed
Oct 8, 2025 at 21:20 UTC
bb6d8c21f9bcc9e333165b81ee99975c64d9ae1d
1 file changed
+2
-2
vllm/v1/engine/detokenizer.py
+2
-2
@@ -227,10 +227,10 @@ class FastIncrementalDetokenizer(BaseIncrementalDetokenizer):
227
def _protected_step(self, next_token_id: int) -> Optional[str]:
228
try:
229
token = self.stream.step(self.tokenizer, next_token_id)
230
- except OverflowError:
230
+ except (OverflowError, TypeError):
231
# Handle rare observed overflow, still to be diagnosed.
232
# See https://github.com/vllm-project/vllm/issues/21951.
233
- logger.exception("Encountered invalid token id: %d", next_token_id)
233
+ logger.exception("Encountered invalid token id: %r", next_token_id)
234
token = None
235
except Exception as e:
236
if not str(e).startswith(INVALID_PREFIX_ERR_MSG):