AI Toolkit adds LoRA training support for music model YuE2-3B
ostrisai · x · 2026-09-15
ostris announced that AI Toolkit, his open-source training tool (12k stars on GitHub), now supports LoRA training for the YuE2-3B music generation model.
- The tokenizer was contributed by @sinceriously (Kytra), with additional help from @Machinedelusion
- The author notes YuE2 is autoregressive and fine-tuning it remains highly experimental
- PR #1042 has been merged into main; training randomly drops sheet-music conditioning 50% of the time so the model works with and without sheet music
More from Multimodal
- AI ads workflow: generate trend-inspired footage with Seedance 2.5, then finish in CapCut — Aiden_Tech_Ai · 2026-09-15
- Reviving a forgotten 1934 French comic weekly with fal, Three.js and GPT 6 Astra — OdinLovis · 2026-09-15
- AI thriller 'All Aboard': 9 generated blocks, color-coded characters on Seedance 2.5 — tsi_org · 2026-09-15
- CMU's SNAP3D turns a single image into physically stable, snap-fit 3D printable parts — anselm · 2026-09-15
- Training LoRAs on Krea2? Skip These txtfusion Layers or Training Explodes — Capitan01R- · 2026-09-15
- HeyGen's open-source HyperFrames hits 50k GitHub stars: agents edit video with HTML — HeyGen · 2026-09-15