@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 64ca424e75
Fork 0

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 17, 06:45 UTC

Fix warning message on LLaMA FastTokenizer (#1037)

Woosuk Kwon committed Sep 14, 2023 at 17:33 UTC 64ca424e7518d1176fff032b76d85e49a4fc936a
1 file changed +5 -5
vllm/transformers_utils/tokenizer.py
+5 -5
@@ -28,8 +28,8 @@ def get_tokenizer(
28 if ("llama" in tokenizer_name.lower() and kwargs.get("use_fast", True)
29 and tokenizer_name != _FAST_LLAMA_TOKENIZER):
30 logger.info(
31 - "For some LLaMA-based models, initializing the fast tokenizer may "
32 - "take a long time. To eliminate the initialization time, consider "
31 + "For some LLaMA V1 models, initializing the fast tokenizer may "
32 + "take a long time. To reduce the initialization time, consider "
33 f"using '{_FAST_LLAMA_TOKENIZER}' instead of the original "
34 "tokenizer.")
35 try:
@@ -41,9 +41,9 @@ def get_tokenizer(
41 except TypeError as e:
42 # The LLaMA tokenizer causes a protobuf error in some environments.
43 err_msg = (
44 - "Failed to load the tokenizer. If you are using a LLaMA-based "
45 - f"model, use '{_FAST_LLAMA_TOKENIZER}' instead of the original "
46 - "tokenizer.")
44 + "Failed to load the tokenizer. If you are using a LLaMA V1 model "
45 + f"consider using '{_FAST_LLAMA_TOKENIZER}' instead of the "
46 + "original tokenizer.")
47 raise RuntimeError(err_msg) from e
48 except ValueError as e:
49 # If the error pertains to the tokenizer class not existing or not