DamRsn/muscriptor-gguf

MuScriptor — GGUF

GGUF conversions of the MuScriptor multi-instrument music transcription weights by Mirelo and Kyutai, converted for use with muscriptor.cpp, a C++/ggml port of the reference implementation, and with NeuralNote.

These are format conversions: safetensors → GGUF, with the bulk weights cast to float16. Nothing was retrained or fine-tuned. All credit for the model belongs to Mirelo and Kyutai — see NOTICE.md.

Files

File	Source checkpoint	Params	Size
v1/muscriptor-small-f16.gguf	muscriptor-small	~103M	200 MiB
v1/muscriptor-medium-f16.gguf	muscriptor-medium	~307M	590 MiB
v1/muscriptor-large-f16.gguf	muscriptor-large	~1.4B	2.6 GiB

medium is the default. small is the one to reach for on a machine without a usable GPU; large is worth it only with one.

The v1/ directory names the checkpoint format generation, which each file also carries as muscriptor.format_version. A loader reads the generation it was built for and refuses the others, so a later generation lands beside this one rather than replacing it, and an application pinned to v1 keeps resolving.

Usage

See muscriptor.cpp. These files are for that library — they are not loadable by llama.cpp or whisper.cpp.

v1/SHA256SUMS lists the hash of every weight file beside it, and each file also carries its own provenance and licence notice as GGUF metadata. Applications embedding these weights should pin a revision hash rather than main, and verify against a checksum compiled into the application rather than one fetched from this repo.

Licence

CC BY-NC 4.0 — non-commercial use only. Full text in LICENSE; deed at https://creativecommons.org/licenses/by-nc/4.0/.

Mirelo and Kyutai supplement the licence with specific conditions of use, reproduced verbatim in NOTICE.md. They are quoted for information; this repository imposes no terms of its own.

Citation

@article{muscriptor2026,
  title   = {MuScriptor: An Open Model for Multi-Instrument Music Transcription},
  author  = {Rouard, Simon and Krause, Michael and Roebel, Axel and
             Simon-Gabriel, Carl-Johann and D{\'e}fossez, Alexandre},
  journal = {arXiv preprint arXiv:2607.08168},
  year    = {2026}
}