lunashia / o-m_beatmap_trainer

What is o-m_beatmap_trainer?

This GitHub training repo delivers a Python-based training pipeline machine learning setup for osu!mania 7K next-event models, predicting note timings and patterns from audio features and past events. It processes beatmaps into supervised datasets, trains PyTorch models on GPU or CPU, and outputs checkpoints ready for audio-driven beatmap generation. Users get a full workflow from raw osu! files to evaluable models, with smoke tests and configurable shards for large datasets.

Why is it gaining traction?

It stands out with a clear training pipeline vs inference pipeline split, handling multimodal inputs like mel spectrograms and grid stats without custom data hacks. The hook is dead-simple CLI scripts for manifest building, cache prep, vocab fitting, and training—perfect for quick baselines in rhythm game AI. Niche focus on 7K mania avoids generic LLM training pipelines, appealing to specialized github training ai experiments.

Who should use this?

Osu! modders prototyping AI beatmap generators from songs. Rhythm game devs needing audio-driven note prediction without starting from scratch. ML hobbyists exploring beatmap training data pipelines, especially if you're opting into github training online for osu! datasets.

Verdict

Grab it if you're in osu! AI—solid baseline pipeline despite low 1.0% credibility from 16 stars and early docs. Run smoke validation first; scale up once your data's ready. Worth forking for custom tweaks.