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 2, 07:15 UTC
[Test][BugFix] Fix double-BOS in PD+specdec acceptance test (#44234)
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Nick Hill committed
Jun 1, 2026 at 14:31 UTC
e4cbc4385d00b4aa817c56de33df30feda13c1f9
1 file changed
+4
tests/v1/kv_connector/nixl_integration/test_spec_decode_acceptance.py
+4
@@ -158,6 +158,10 @@ def test_spec_decode_acceptance_length():
158
max_tokens=DEFAULT_OUTPUT_LEN,
159
temperature=0.0,
160
top_p=1.0,
161
+ # Prompts are already chat-templated (contain BOS); avoid the
162
+ # completions API prepending a second BOS, which would lower
163
+ # acceptance ~5% vs the add_special_tokens=False standalone baselines.
164
+ extra_body={"add_special_tokens": False},
165
)
166
if i < 3:
167
text = resp.choices[0].text.strip()[:100]