Spaces:
Runtime error
Runtime error
Commit History
-
Fix HF Space cold-start timeout: slim deps + prebake small model
Shubham32142 +1: Claude Opus 4.7 (1M context)- Drop torch/torchaudio/transformers/datasets from runtime requirements; faster-whisper uses CTranslate2, not PyTorch. Moved training-only deps to ml/finetune-requirements.txt. - Pre-download Systran/faster-whisper-small during image build so the first /transcribe request doesn't trigger a multi-minute download on the free CPU tier. - Default MODEL_PATH to the HF model id instead of a local ../models path. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>a2fed52
-
Remove large/medium models, support only tiny/base/small; exclude models from git
Shubham32142d344109 -
Add Docker support and implement Whisper transcription service
Shubham32142- Created .dockerignore to exclude unnecessary files from Docker context. - Added Dockerfile for the ML service, installing dependencies and setting up the environment. - Created Dockerfile for the ML model, including necessary installations. - Added docker-compose.yml to orchestrate ML and API services with environment variables. - Introduced .env file for ML service configuration. - Added config.yaml for model configuration parameters. - Implemented evaluation script for measuring WER/CER in evaluate.py. - Created dataset preparation script in prepare_dataset.py for fine-tuning. - Added script to push fine-tuned models to Hugging Face Hub in push_to_hub.py. - Implemented training script for fine-tuning Whisper model in train.py. - Updated requirements.txt with necessary dependencies for ML service. - Developed serve.py to handle transcription requests via FastAPI. - Created standalone transcriber script in transcribe.py. - Added model directory structure with necessary files for Whis…a4a3878