Instructions to use moonbox-dev/heartmula with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use moonbox-dev/heartmula with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download moonbox-dev/heartmula --local-dir heartmula
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
HeartMuLa 3B for Moonbox
The files the Moonbox desktop app downloads for these models:
- HeartMuLa 3B (HeartMuLa): Songs with sung vocals from your lyrics and a style, up to four minutes.
They are the HeartMuLa 3B music model converted to MLX for Apple silicon (8-bit for the app; 6-bit, mixed and bf16 builds are here too) and the HeartCodec decoder. Moonbox checks every file against a signed list before using it. You do not need this repository to use Moonbox, and it is not an official release of the original models.
What is here
| Path | What | From |
|---|---|---|
models/mlx/HeartMuLa-3B-*, models/hf/HeartMuLaGen/.gitattributes, models/hf/HeartMuLaGen/README.md, models/hf/HeartMuLaGen/gen_config.json |
HeartMuLa-oss-3B-happy-new-year LM (8-bit required; 6-bit, mixed 4-bit and bf16 optional) + gen config | https://huggingface.co/HeartMuLa/HeartMuLa-oss-3B-happy-new-year |
models/mlx/HeartCodec-decoder.safetensors |
HeartCodec-oss-20260123 decoder (MLX) | https://huggingface.co/HeartMuLa/HeartCodec-oss-20260123 |
models/hf/HeartMuLaGen/tokenizer.json |
HeartMuLaGen tokenizer (Meta's Llama 3.2 tokenizer) | https://huggingface.co/HeartMuLa/HeartMuLaGen (tokenizer.json, identical to Meta's Llama 3.2 tokenizer) |
models/torch/heartmula-rl-int8wo/, models/torch/heartcodec-bf16/ |
an 8-bit weight-only export of the HeartMuLa language model and a bf16 HeartCodec decoder, for the Windows engine | https://huggingface.co/HeartMuLa |
Every part was converted, quantized or repackaged from its source unless its row says "as published": treat none of these files as the original release.
Licences
Built with Llama. The model and the codec are Apache-2.0. The tokenizer (models/hf/HeartMuLaGen/tokenizer.json) is Meta's Llama 3.2 tokenizer: Llama 3.2 is licensed under the Llama 3.2 Community License, Copyright ยฉ Meta Platforms, Inc. All Rights Reserved, and its use is subject to the Acceptable Use Policy at https://www.llama.com/llama3_2/use-policy. That one file is why this repository declares that licence; the full text is in LICENSE.
| Part | Licence | Notice |
|---|---|---|
| HeartMuLa-oss-3B-happy-new-year LM (8-bit required; 6-bit, mixed 4-bit and bf16 optional) + gen config | Apache License 2.0 | HeartMuLa authors |
| HeartCodec-oss-20260123 decoder (MLX) | Apache License 2.0 | HeartMuLa authors |
| HeartMuLaGen tokenizer (Meta's Llama 3.2 tokenizer) | Llama 3.2 Community License Agreement; Apache License 2.0 | Built with Llama. Llama 3.2 is licensed under the Llama 3.2 Community License, Copyright ยฉ Meta Platforms, Inc. All Rights Reserved. Use of Llama Materials is subject to the Acceptable Use Policy: https://www.llama.com/llama3_2/use-policy |
| an 8-bit weight-only export of the HeartMuLa language model and a bf16 HeartCodec decoder, for the Windows engine | Apache License 2.0 |
The full text of each licence is in LICENSE. The models are the work of their authors, linked above. Moonbox converted and packaged them and claims no rights in them. Use is subject to each licence above.
Quantized
Model tree for moonbox-dev/heartmula
Base model
HeartMuLa/HeartCodec-oss-20260123