CoolFace
Modelpublic

mmmino/TrackLM

sourceHugging Facellama3.2updated 17d agoView on Hugging Face
0likes
Model Card

TrackLM

Two-stage text-to-MIDI model built on Llama 3.2 1B with an extended vocabulary (128,256 text tokens + 55,030 music event tokens <|ext_0|> … <|ext_55029|>).

  • —planner/ — caption → bar-by-bar plan (text)
  • —renderer/ — plan → music event tokens

Both are LlamaForCausalLM checkpoints in bf16, loadable with transformers.

Built with Llama.