Show HN: EDM – pick a track and a transformer generates a DJ set, track-by-track (electronicdjmodel.com)

The Electronic DJ Model (EDM) has been launched, showcasing a transformer model that autonomously generates DJ playlists based on a user's selected track. This innovative model was trained on a dataset called the DJ Mix Dataset, containing 5,040 real DJ sets which allow it to curate a personalized mix for users in a coherent track-by-track format. By utilizing a novel tokenizer to manage 63,040 unique tracks, EDM stands out for its ability to create mixes that reflect learned patterns rather than simple repetitions, with a smaller ~8 million-parameter model outperforming larger counterparts. This development is significant for the AI/ML community as it demonstrates the potential of transformer models in creative applications like music curation. The model not only showcases technical advancements in natural language processing techniques applied to audio sequencing but also incorporates user feedback via a thumbs-up and thumbs-down system. This feedback mechanism aids in model improvement through Direct Preference Optimization (DPO), further pushing the boundaries of AI in entertainment and music industries.