Generate melodies with transformers!
This project implements a melody generation pipeline using transformer models. It includes scripts for preprocessing MIDI data, training a transformer model, and generating new melodies based on a given starting sequence. The implementation leverages PyTorch for the model, Mido for MIDI file processing, and DVC for data and model versioning.
- Data and Model Versioning: Utilizes DVC to track datasets and models
- Transformer Model: Custom transformer architecture tailored for sequence-to-sequence melody generation
- Inference Pipeline: Scripts to generate melodies and save them as MIDI files
- BentoML service for serving models in production
- Prometheus for monitoring and alerting
- Containerization using Docker for consistent deployments
- GitHub Action workflows for seamless CI/CD integration with AWS and Docker