Missing model card and configuration
Hi Alex, thanks for sharing Music Llama 4,
I’m trying to run the published checkpoint locally. Could you point me to its matching inference notebook or source code?
Specifically, I’m looking for:
The complete model configuration and required Transformers version.
The mapping between the 1,028 token IDs and musical events, including special tokens.
MIDI encoding/decoding code.
An example of constructing a valid prompt and generating a MIDI file.
Are prompts intended to be seed MIDIs/event tokens, or does this model support text prompts?
@korigamikThank you for writing and for your interest in the project.
Please see the link below for project files archive which includes everything you asked for.
https://huggingface.co/datasets/asigalov61/misc-and-temp/blob/main/Music-Llama-4-Project-Files.zip
Music LLaMa 4 is a pure MIDI model, not text to MIDI model. Text to MIDI is very hard to do due to the fact that there are not properly annotated and sufficiently large datasets available still.
This project was inspired by https://huggingface.co/spaces/skytnt/midi-composer which was my colab project with SkyTNT.
Also, consider using x-transformers by lucidrains. LLaMa architecture is not very well suited for music.
I sincerely apologize for the messy codebase due to the fact that this project was never finished and because I depreciated it some time ago.
However, feel free to use/adapt it as you wish.
License is Aparche 2.0 for code and CC BY-NC-SA 4.0 for artwork.
Most sincerely,
Alex.
P.S. I highly recommend you check out my current projects here: https://huggingface.co/projectlosangeles