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 8, 20:15 UTC
Fix tokenizer test for renamed attr on Transformers v5 (#33902)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Feb 5, 2026 at 19:16 UTC
1887acca9e2ceacea8f7b1770bd0a0fd9b6a3b02
1 file changed
+9
-1
tests/entrypoints/openai/test_serving_tokens.py
+9
-1
@@ -7,6 +7,7 @@ import pytest_asyncio
7
from transformers import AutoTokenizer
8
9
from vllm.config import ModelConfig
10
+from vllm.config.utils import getattr_iter
11
from vllm.v1.engine.detokenizer import check_stop_strings
12
13
from ...utils import RemoteOpenAIServer
@@ -131,7 +132,14 @@ async def test_same_response_as_chat_completions(client, tokenizer, messages):
132
# Post-EOS generation is undefined and may differ
133
eos_tokens = {
134
tokenizer.eos_token_id,
134
- *tokenizer.additional_special_tokens_ids,
135
+ *getattr_iter(
136
+ tokenizer,
137
+ [
138
+ "extra_special_tokens_ids", # Transformers v5
139
+ "additional_special_tokens_ids", # Transformers v4
140
+ ],
141
+ [],
142
+ ),
143
}
144
# Find first EOS in generated tokens
145
eos_pos = None