Spaces:
Running
on
Zero
Running
on
Zero
| # 1) Do NOT pin torch/torchaudio here — use the CUDA builds that come with the image | |
| transformers==4.48.0 | |
| # Removed flash-attention since faster-whisper handles this internally | |
| # https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.0.8/flash_attn-2.7.4.post1+cu126torch2.4-cp310-cp310-linux_x86_64.whl | |
| pydantic==2.10.6 | |
| # 2) Main whisper stack — WhisperX (with compatible FW/CT2 for cuDNN 8 era) | |
| whisperx==3.7.0 | |
| faster-whisper==1.2.0 | |
| ctranslate2==4.5.0 | |
| # (no 'torch' line — keep the image’s CUDA wheel) | |
| # Optional: stable-ts for regroup/edit post-processing | |
| stable-ts>=2.13.3 | |
| # 3) Extra libs your app really needs | |
| gradio==5.0.1 | |
| spaces>=0.19.0 | |
| pyannote.audio>=3.3.1 | |
| pandas>=1.5.0 | |
| numpy>=1.24.0 | |
| librosa>=0.10.0 | |
| soundfile>=0.12.0 | |
| ffmpeg-python>=0.2.0 | |
| requests>=2.28.0 | |
| webrtcvad>=2.0.10 | |
| boto3 | |