Requirements.txt Comparison Analysis
Critical Missing Dependencies in HF requirements.txt
1. Pyannote Specific Versions (CRITICAL)
The HF requirements.txt has pyannote.audio>=3.1.1 but the working version uses specific pinned versions:
pyannote.audio==3.4.0
pyannote.core==5.0.0
pyannote.database==5.1.3
pyannote.metrics==3.2.1
pyannote.pipeline==3.0.1
pyannoteai-sdk==0.3.0
2. OpenAI Version Mismatch (CRITICAL)
- HF:
openai>=1.50.0 - Working:
openai>=2.0.0,<3.0.0(actual: 2.8.1)
3. Pinecone Version Mismatch (CRITICAL)
- HF:
pinecone-client>=5.0.0 - Working:
pinecone-client>=6.0.0(actual: 6.0.0)
4. Tiktoken Version Mismatch
- HF:
tiktoken>=0.8.0 - Working:
tiktoken>=0.10.0,<1.0.0(actual: 0.12.0)
5. Missing Dependencies (Not in HF requirements.txt)
These packages are installed in the working venv but not specified in HF requirements:
lightning==2.5.6pytorch-lightning==2.5.6pytorch-metric-learning==2.9.0speechbrain==1.0.3sentencepiece==0.2.1transformers==4.57.3tokenizers==0.22.1tensorboardX==2.6.4einops==0.8.1omegaconf==2.3.0HyperPyYAML==1.2.2asteroid-filterbanks==0.4.0julius==0.2.7onnxruntime==1.23.2protobuf==6.33.1coloredlogs==15.0.1humanfriendly==10.0
6. Torch Audio Processing Dependencies
torch-audiomentations==0.12.0torch_pitch_shift==1.2.5torchcodec==0.7.0torchmetrics==1.8.2
7. Additional Data Processing
av==15.1.0(for video processing)imageio==2.37.2imageio-ffmpeg==0.6.0moviepy==2.2.1pydub==0.25.1
Recommendations
High Priority (MUST FIX)
- Pin pyannote versions exactly as in working environment
- Update OpenAI to 2.x range
- Update Pinecone to 6.x
- Add transformers, tokenizers, sentencepiece (likely needed for model loading)
- Add speechbrain (likely needed by pyannote)
Medium Priority (SHOULD ADD)
- Add PyTorch ecosystem packages: pytorch-lightning, pytorch-metric-learning, torchmetrics
- Add audio processing: torch-audiomentations, torch_pitch_shift, torchcodec
- Add video processing: av, moviepy, imageio
Low Priority (OPTIONAL)
- Add logging utilities: coloredlogs, humanfriendly
- Add config management: omegaconf, HyperPyYAML