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
Revert "[PERF] Use faster way of decode in tokenizer: avoid useless list-to-list conversion (#20000)" (#23396)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Aug 23, 2025 at 12:16 UTC
b4e9fd811f528ae54e620147ae9e2a09af50358f
1 file changed
+3
-4
vllm/transformers_utils/tokenizer.py
+3
-4
@@ -49,12 +49,11 @@ def decode_tokens(
49
`skip_special_tokens=None` means to use the backend's default
50
settings.
51
"""
52
- decode_method = getattr(tokenizer, "_decode", tokenizer.decode)
52
if skip_special_tokens is not None:
54
- return decode_method(token_ids,
55
- skip_special_tokens=skip_special_tokens)
53
+ return tokenizer.decode(token_ids,
54
+ skip_special_tokens=skip_special_tokens)
55
57
- return decode_method(token_ids)
56
+ return tokenizer.decode(token_ids)
57
58
59
def encode_tokens(