demucs-htdemucs-mlx (HTDemucs v4, MLX)

MLX conversion of facebookresearch/demucs for inference on Apple silicon via mlx-audiogen.

Licence

Upstream Demucs is MIT — both the code and the released checkpoints, per the project's LICENSE and its README. MIT is permissive, including for commercial use, but it is not obligation-free: the licence requires that

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

which is why this repository ships the full LICENSE rather than only a license: mit tag.

mlx-audiogen's own Apache-2.0 licence covers the conversion code only. The full licence is in LICENSE; attribution and the statement of modifications are in NOTICE.

Attribution

HTDemucs v4 was created by Simon Rouard, Francisco Massa and Alexandre Défossez, and published in Hybrid Transformers for Music Source Separation.

Modifications

These weights are a modified version of Meta's HTDemucs v4 checkpoint, converted to the MLX array format for inference on Apple silicon. Specifically: the upstream pickled Torch .th checkpoint is read for its tensor state dict only, weight keys are remapped to match this project's module names, convolution weights are transposed where MLX's layout differs from PyTorch's (both the 1-D and 2-D cases), fused QKV projections are split into separate Q, K and V tensors, and the result is re-serialised as safetensors. The model is re-implemented against MLX rather than PyTorch.

No weights are retrained or fine-tuned.

Tensor precision is unchanged: the upstream checkpoint stores float32 and so does this conversion. No weight value is altered by rounding — the Demucs converter performs no dtype cast.

Usage

from mlx_audiogen.models.demucs.pipeline import DemucsPipeline

# Auto-downloads weights from this repo
pipeline = DemucsPipeline.from_pretrained()

stems = pipeline.separate(audio_np, sample_rate=44100)
# stems = {"drums": np.ndarray, "bass": ..., "other": ..., "vocals": ...}

The weights download automatically on first use.

Provenance

The checkpoint this was converted from is distributed by Meta at https://dl.fbaipublicfiles.com/demucs/hybrid_transformer/955717e8-8726e21a.th, referenced by the upstream project above.