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 16, 19:30 UTC
[Bugfix] fix logit processor excceed vocab size issue (#6927)
Fei committed
Jul 31, 2024 at 01:16 UTC
c0644cf9cef0002485749defcaa02e3fec359d49
1 file changed
+6
vllm/entrypoints/openai/logits_processors.py
+6
@@ -58,6 +58,12 @@ def get_logits_processors(
58
"Found token_id in logit_bias that is not "
59
"an integer or string representing an integer") from exc
60
61
+ # Check if token_id is within the vocab size
62
+ for token_id, bias in clamped_logit_bias.items():
63
+ if token_id < 0 or token_id >= tokenizer.vocab_size:
64
+ raise ValueError("token_id in logit_bias contains "
65
+ "out-of-vocab token id")
66
+
67
def logit_bias_logits_processor(token_ids: List[int],
68
logits: torch.Tensor) -> torch.Tensor:
69
for token_id, bias in clamped_logit_bias.items():