youngkingdom/vllm
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[VLM] Merged multimodal processor for Qwen2-Audio (#11303)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2024-12-19 14:14:17 +08:00
committed by GitHub
parent c6b0a7d3ba
commit 6142ef0ada
11 changed files with 414 additions and 358 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
tests/models/decoder_only/audio_language/test_ultravox.py
View File

@ -5,6 +5,7 @@ import pytest
import pytest_asyncio
from transformers import AutoModel, AutoTokenizer, BatchEncoding
from vllm.multimodal.audio import resample_audio
from vllm.sequence import SampleLogprobs
from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE
@ -130,16 +131,14 @@ def run_test(
dtype=dtype,
postprocess_inputs=process,
auto_cls=AutoModel) as hf_model:
import librosa
hf_outputs_per_audio = [
hf_model.generate_greedy_logprobs_limit(
[hf_prompt],
max_tokens,
num_logprobs=num_logprobs,
audios=[(librosa.resample(audio[0],
orig_sr=audio[1],
target_sr=16000), 16000)])
audios=[(resample_audio(audio[0],
orig_sr=audio[1],
target_sr=16000), 16000)])
for _, hf_prompt, audio in prompts_and_audios
]