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 5, 21:00 UTC
[CI] Add temperature=0.0, reduce max_tokens, and add debug prints to audio_in_video tests (#38492)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Mar 30, 2026 at 00:36 UTC
bea23536f627b2b4153f2e672753b6034b78dedb
1 file changed
+18
-4
tests/entrypoints/openai/chat_completion/test_audio_in_video.py
+18
-4
@@ -64,11 +64,12 @@ async def test_online_audio_in_video(
64
]
65
66
# multi-turn to test mm processor cache as well
67
- for _ in range(2):
67
+ for turn in range(2):
68
chat_completion = await client.chat.completions.create(
69
model=MODEL_NAME,
70
messages=messages,
71
- max_tokens=16,
71
+ max_tokens=8,
72
+ temperature=0.0,
73
extra_body={
74
"mm_processor_kwargs": {
75
"use_audio_in_video": True,
@@ -78,6 +79,12 @@ async def test_online_audio_in_video(
79
80
assert len(chat_completion.choices) == 1
81
choice = chat_completion.choices[0]
82
+ print(
83
+ f"[DEBUG][single-video] turn={turn} "
84
+ f"finish_reason={choice.finish_reason!r} "
85
+ f"content={choice.message.content!r} "
86
+ f"usage={chat_completion.usage}"
87
+ )
88
assert choice.finish_reason == "length"
89
90
@@ -111,11 +118,12 @@ async def test_online_audio_in_video_multi_videos(
118
]
119
120
# multi-turn to test mm processor cache as well
114
- for _ in range(2):
121
+ for turn in range(2):
122
chat_completion = await client.chat.completions.create(
123
model=MODEL_NAME,
124
messages=messages,
118
- max_tokens=16,
125
+ max_tokens=8,
126
+ temperature=0.0,
127
extra_body={
128
"mm_processor_kwargs": {
129
"use_audio_in_video": True,
@@ -125,6 +133,12 @@ async def test_online_audio_in_video_multi_videos(
133
134
assert len(chat_completion.choices) == 1
135
choice = chat_completion.choices[0]
136
+ print(
137
+ f"[DEBUG][multi-video] turn={turn} "
138
+ f"finish_reason={choice.finish_reason!r} "
139
+ f"content={choice.message.content!r} "
140
+ f"usage={chat_completion.usage}"
141
+ )
142
assert choice.finish_reason == "length"
143
144