xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 07:24:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

[tests] Improve speed and reliability of test_transcription_api_correctness (#23854)

Browse Source 
Signed-off-by: Russell Bryant <rbryant@redhat.com>
This commit is contained in:
Russell Bryant 2025-08-29 00:25:33 -04:00 committed by GitHub
parent 006477e60b
commit c8b3b299c9
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
tests/entrypoints/openai/correctness/test_transcription_api_correctness.py
View File

@ -49,8 +49,7 @@ async def transcribe_audio(client, tokenizer, y, sr):
return latency, num_output_tokens, transcription.text
async def bound_transcribe(model_name, sem, client, audio, reference):
tokenizer = AutoTokenizer.from_pretrained(model_name)
async def bound_transcribe(sem, client, tokenizer, audio, reference):
# Use semaphore to limit concurrent requests.
async with sem:
result = await transcribe_audio(client, tokenizer, *audio)
@ -63,15 +62,19 @@ async def bound_transcribe(model_name, sem, client, audio, reference):
async def process_dataset(model, client, data, concurrent_request):
sem = asyncio.Semaphore(concurrent_request)
# Load tokenizer once outside the loop
tokenizer = AutoTokenizer.from_pretrained(model)
# Warmup call as the first `librosa.load` server-side is quite slow.
audio, sr = data[0]["audio"]["array"], data[0]["audio"]["sampling_rate"]
_ = await bound_transcribe(model, sem, client, (audio, sr), "")
_ = await bound_transcribe(sem, client, tokenizer, (audio, sr), "")
tasks: list[asyncio.Task] = []
for sample in data:
audio, sr = sample["audio"]["array"], sample["audio"]["sampling_rate"]
task = asyncio.create_task(
bound_transcribe(model, sem, client, (audio, sr), sample["text"]))
bound_transcribe(sem, client, tokenizer, (audio, sr),
sample["text"]))
tasks.append(task)
return await asyncio.gather(*tasks)