Add initial validated seven-model MLX audio separation catalog

Publish safetensors, matching configurations, model cards, validation reports, checksums, and catalog-specific third-party notices. Automatic downloads require the upcoming mlx-audio-separator 0.1.21 release.

.gitattributes

@@ -1,35 +1 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
 *.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text

LICENSE-code

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2024-2026 ssmall256
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -0,0 +1,193 @@
+---
+library_name: mlx
+pipeline_tag: audio-to-audio
+license: unknown
+tags:
+- mlx
+- audio
+- music-source-separation
+- apple-silicon
+- safetensors
+---
+
+# MLX Audio Separator
+
+Separate music into vocals, drums, bass, guitar, piano, and other instruments on your
+Apple Silicon Mac. This repository contains seven ready-to-use **safetensors models** for
+[mlx-audio-separator](https://github.com/ssmall256/mlx-audio-separator), with MLX inference,
+matching configurations, and individual validation reports. No PyTorch or ONNX runtime is
+required to use these weights.
+
+Download one model or the full collection. All weights are hosted here, organized by model ID.
+
+## Models
+
+| Model | Output stems | Precision | Size |
+|---|---|---|---:|
+| [BS-RoFormer-SW](cards/bs-roformer-sw.md) | Bass, drums, other, vocals, guitar, piano | float32 | 666.5 MiB |
+| [Mel-RoFormer InstV7n](cards/mel-roformer-instv7n.md) | Instrumental, vocals | float32 | 870.6 MiB |
+| [UVR-MDX-NET Inst HQ3](cards/uvr-mdx-net-inst-hq3.md) | Instrumental, vocals | float32 | 63.8 MiB |
+| [HTDemucs](cards/htdemucs.md) | Drums, bass, other, vocals | float32 | 160.2 MiB |
+| [HTDemucs_FT](cards/htdemucs_ft.md) | Drums, bass, other, vocals | float32 | 640.9 MiB |
+| [HTDemucs_6s](cards/htdemucs_6s.md) | Drums, bass, other, vocals, guitar, piano | float32 | 104.6 MiB |
+| [ZFTurbo Mel-RoFormer Vocals v1](cards/mel-roformer-zfturbo-vocals-v1.md) | Vocals, instrumental | float16 | 64.3 MiB |
+
+**BS-RoFormer-SW** separates six instruments in one model. **HTDemucs** produces four stems;
+**HTDemucs_FT** uses four specialist models, and **HTDemucs_6s** adds guitar and piano.
+For vocal and instrumental separation, choose InstV7n, Inst HQ3, or ZFTurbo Vocals v1.
+
+All models produce stereo audio at **44.1 kHz**. The complete collection is **2.51 GiB**.
+Download sizes include weights and configurations; runtime memory use is higher and varies by model.
+Weight precision describes the stored tensors, not the precision of every inference operation.
+
+## Quick start
+
+Requires an **Apple Silicon Mac running macOS 26+**, native **Python 3.10+**, and
+**mlx-audio-separator 0.1.21 or later**.
+
+**Release availability:** automatic downloads require the upcoming 0.1.21 package release.
+The model files are available now through [direct downloads](#direct-downloads).
+The CLI and Python examples below apply once 0.1.21 is released; PyPI 0.1.20 does not
+include this catalog integration.
+
+Install [mlx-audio-separator](https://github.com/ssmall256/mlx-audio-separator) and choose a model:
+
+```bash
+python -m pip install "mlx-audio-separator>=0.1.21"
+mlx-audio-separator song.m4a -m bs-roformer-sw --output_dir stems
+```
+
+**The first use downloads the selected model automatically.** The package verifies its weights
+and configuration against bundled SHA-256 hashes, caches them, and separates your audio.
+Later runs reuse the cache and work offline. You do not need to download or convert a checkpoint yourself.
+
+The CLI writes FLAC by default; add `--output_format WAV` for WAV files.
+
+### Python
+
+```python
+from mlx_audio_separator import Separator
+
+separator = Separator(output_dir="stems", output_format="WAV")
+separator.load_model("bs-roformer-sw")
+output_files = separator.separate("song.m4a")
+```
+
+Reuse the separator for additional files. Each of the seven model cards lists its short catalog ID
+and existing package name; both select the same safetensors. For example, `BS-Roformer-SW.ckpt`
+remains an alias for `bs-roformer-sw`, and `htdemucs_ft.yaml` remains an alias for `htdemucs_ft`.
+The `.ckpt`, `.onnx`, and `.yaml` aliases are model names, not the formats downloaded for these entries.
+Choose a model explicitly: the package's historical default is a different BS-RoFormer checkpoint.
+
+## Cache and offline use
+
+Automatic downloads go to `~/.cache/mlx-audio-separator/models/<model-id>/`.
+Use `--model_file_dir /path/to/models` in the CLI, or `model_file_dir="/path/to/models"` in Python,
+to choose another model directory. `AUDIO_SEPARATOR_MODEL_DIR` can select an existing directory
+for both interfaces and takes precedence over those options.
+
+Set `HF_HUB_OFFLINE=1` or `MLX_AUDIO_SEPARATOR_NO_DOWNLOAD=1` to prohibit downloads.
+A complete, verified cache works without network access. A missing or damaged file produces a
+clear error in offline mode; with downloads enabled, the package fetches the required replacement.
+Failed transfers do not overwrite existing files.
+
+To free space, remove an unused model's folder. The next use restores the required files.
+The package also reuses matching files from older model caches where possible.
+
+## Direct downloads
+
+You can also download weights and configurations directly from this repository's **Files and
+versions** tab, or use the Hugging Face client. This is useful for manual setup or preparing
+an offline machine.
+
+```bash
+python -m pip install huggingface_hub
+```
+
+```python
+from huggingface_hub import snapshot_download
+
+snapshot_download(
+    repo_id="ssmall256/mlx-audio-separator",
+    local_dir="mlx-separator-models",
+    allow_patterns=["models/bs-roformer-sw/*"],
+)
+```
+
+Point the normal CLI at the downloaded `models/` directory:
+
+```bash
+HF_HUB_OFFLINE=1 mlx-audio-separator song.m4a \
+  -m bs-roformer-sw \
+  --model_file_dir mlx-separator-models/models \
+  --output_dir stems
+```
+
+Keep each model's files together under its model ID. Replace `bs-roformer-sw` in both examples
+to choose another model, or use `models/*` as the download pattern for all seven.
+The package verifies manually downloaded files using the same checks as automatic downloads.
+
+For reproducible downloads, pass a specific commit with `snapshot_download(revision=...)`.
+Use a package release whose bundled file hashes match that revision. See the
+[Hugging Face download guide](https://huggingface.co/docs/huggingface_hub/guides/download) for more options.
+
+## Inference settings
+
+RoFormer models use **two overlapping windows** by default: each chunk starts halfway through
+the preceding chunk. Demucs uses **two shifts** and **25% overlap**. Shift offsets are random
+by default; set `--demucs_seed 0` or `demucs_params={"seed": 0}` for repeatable selection.
+Inst HQ3 uses 25% overlap and a segment size of 256. The normal package options remain available
+to adjust these settings.
+
+The [catalog manifest](catalog.json) records filenames, sizes, SHA-256 hashes, source checkpoints,
+and conversion details. Each model card includes file hashes, and `SHA256SUMS` covers the
+repository's distributed files.
+
+## Validation
+
+Every model passed offline inference on a 14-second real-music excerpt in a fresh process,
+producing the expected full-length, finite stereo stems at 44.1 kHz. These checks ran with
+network access disabled and imports of PyTorch, ONNX, and ONNX Runtime blocked.
+
+The BS-RoFormer-SW, InstV7n, and Inst HQ3 conversions produced **bit-identical unquantized stems**
+to their original checkpoints loaded through the same MLX implementations. The three Demucs
+models and ZFTurbo Vocals v1 preserve the bytes of their existing published safetensors,
+verified against pinned source revisions.
+
+The normal package loader also passed first-use download and offline-cache checks for all seven
+models using a local HTTP mirror of these files. Both runs produced bit-identical unquantized
+stems, with conversion dependencies blocked. See the [automatic-download report](validation/automatic-downloads.json).
+
+These checks establish artifact integrity and conversion consistency. The 14-second check is
+not a full quality evaluation or a speed benchmark. See [Validation details](VALIDATION.md)
+for per-model reports and earlier comparisons against
+[python-audio-separator](https://github.com/nomadkaraoke/python-audio-separator), the upstream
+project used for PyTorch references.
+
+## Licensing and credit
+
+Model weights have individual licensing terms; there is no single license covering the entire
+collection. The **MIT** license in [LICENSE-code](LICENSE-code) applies to the catalog helper.
+See [Third-Party Notices](THIRD_PARTY_NOTICES.md) for checkpoint and conversion credits,
+upstream software attribution, and links to the included model-specific license texts.
+
+| Models | Weight license information |
+|---|---|
+| BS-RoFormer-SW | The [source model card](https://huggingface.co/enerjazzer/BS-ROFO-SW-Fixed/tree/a443a2985534b3bc815ef54a5d446c6a0390f974) lists the license as **unknown**. |
+| Mel-RoFormer InstV7n | No weight-specific redistribution license is recorded in the source metadata. |
+| UVR-MDX-NET Inst HQ3 | No weight-specific redistribution license is recorded in the source metadata. |
+| HTDemucs, HTDemucs_FT, HTDemucs_6s | **MIT**, as declared by the pinned [Demucs MLX source repository](https://huggingface.co/ssmall256/demucs-mlx/tree/35ecb1a906ec420b1a28069db575ad070d35860e); license notices accompany the weights. |
+| ZFTurbo Vocals v1 | The [source card](https://huggingface.co/mlx-community/mel-roformer-zfturbo-vocals-v1-mlx/tree/01bef2daebedf39937f552d6da68340fc979c9a0) declares **MIT**, with a caveat that the source license was added after the checkpoint release. Its notice is included unchanged. |
+
+An unknown weight license is not a grant of permission. Consult the individual model cards and
+original sources for the terms relevant to your use.
+
+The fixed BS-RoFormer-SW checkpoint traces to **jarredou**'s distribution, mirrored by **enerjazzer**;
+InstV7n to **Gabox**; Inst HQ3 to the **UVR/TRvlvr** model release; Demucs to **Meta and the Demucs
+contributors**; and ZFTurbo Vocals v1 to **Roman Solovyev (ZFTurbo)**, with its existing MLX conversion
+from **mlx-community**. Architecture credit belongs to the BS-RoFormer, Mel-Band-RoFormer,
+Demucs, and MDX/UVR authors and contributors.
+
+MLX separator development and catalog preparation are maintained by
+[ssmall256](https://huggingface.co/ssmall256). Report runtime issues or request additional models
+in the [mlx-audio-separator issue tracker](https://github.com/ssmall256/mlx-audio-separator/issues).

SHA256SUMS

@@ -0,0 +1,40 @@
+ee05350c7fda0af4da116020dd7b976d731de3beed4f5e717d5bcc95ead24550  .gitattributes
+980ed5199ba2b41c564a024ef183f29cec371c1d77b65982a8e21c037bee74f6  LICENSE-code
+ac88c30e525c3f634bb5e16fb00da514692c2aff84975c7e8fd37826db7b42ed  README.md
+3685afee0763ef6c3be1718e425dda024dbc69dd01cb82122579eb689b3e80d2  THIRD_PARTY_NOTICES.md
+0946bc24fa11497c3b2a302a96d2bc21cb8dcfa09cb501a1f70f2e34d33c3139  VALIDATION.md
+bb2e30830573d781821f62ae5c7415e021863a99cf1eca67df1e81e42b41b4b3  cards/bs-roformer-sw.md
+da22bc2b2fbc749e666c57b471cc7273ec799b02f7525cb3f2a2c5d674ebc3b5  cards/htdemucs.md
+d926563e4b56b6cf25dcd0e720b5297d81cf14f4fcc63034a8ff3aed54ebba7f  cards/htdemucs_6s.md
+740196e8faba91f573bceae68e62c1c7d959c91c3ed66f9731790cf265395e03  cards/htdemucs_ft.md
+e5edfd2544b3ae2a2202303062298f0c67e7b4df38eb47050afcb0dd28d165fa  cards/mel-roformer-instv7n.md
+b469ee5f16282aef423f28d6299fd961a04bf729c04f9cb6877760d6c286a284  cards/mel-roformer-zfturbo-vocals-v1.md
+842bf30a9b310b2c8c445d20002ec0b907b454c00811fc202793ce5dc52c6b4a  cards/uvr-mdx-net-inst-hq3.md
+8d26f8b48adfe3c96e60f1a360e30eecfa24b620539cd55714d22012ec8dbe95  catalog.json
+41469a84c9cd1ea89be7cc98eae2e72837e5b8d897b9d3fc368d1538015a5c14  catalog.py
+386a86c2061536d2dca349943f1027f3ac5472ccb22b0e18bd528519b90f8931  models/bs-roformer-sw/config.json
+f95907cd5bfdbc783c1cb32f6aafc49b8b6a598477a9d4a48b4f9191c5167707  models/bs-roformer-sw/model.safetensors
+15086279d32c0f00c577c0f52ff428daf98b8a1fec0264da1c717c88ad464f51  models/htdemucs/LICENSE
+339d267a7a6983a11eedbdc00413c602a65e9b9103f695fb5c2b2a481cd9d297  models/htdemucs/htdemucs.safetensors
+23657b19db14771aecf366ceedbf846c1386836c10043c97f4569acf05526b78  models/htdemucs/htdemucs_config.json
+15086279d32c0f00c577c0f52ff428daf98b8a1fec0264da1c717c88ad464f51  models/htdemucs_6s/LICENSE
+d298f7f746bf53c21baad44fb08e88807ef47feb551dd22f1601a546c85b8e02  models/htdemucs_6s/htdemucs_6s.safetensors
+d5e18c0209be583027d6eb5ec3013a02cd7ef55bc3096e644d8d3577d92e9bd0  models/htdemucs_6s/htdemucs_6s_config.json
+15086279d32c0f00c577c0f52ff428daf98b8a1fec0264da1c717c88ad464f51  models/htdemucs_ft/LICENSE
+53f03b1ad4b4d211025a35da65460ba61a17547adf9c0544cad0ebcc8d7bbabb  models/htdemucs_ft/htdemucs_ft.safetensors
+6708535a698d4d4dd285a6f0a80d9d3c228fdd6514e53bdf1ddbc2ee9b0fa872  models/htdemucs_ft/htdemucs_ft_config.json
+cd8548fd6a1345cfd0c51970919e15e7262cee954b6ff45bb016ed1364be38c5  models/mel-roformer-instv7n/config.json
+74397de5395ed372059b971d6398106f60d92373ada69b62e986e52c9697a5ba  models/mel-roformer-instv7n/model.safetensors
+2b55ae02dcfe9ac8454c41c5e235175512db7429ee7275bab4c4e566b3f80597  models/mel-roformer-zfturbo-vocals-v1/LICENSE
+d110ed4713b403d27e2ae6c7796b1a864576232de4a2070e9762108ca197caa6  models/mel-roformer-zfturbo-vocals-v1/config.json
+ef4aa052845a868cfaff93611477bd8f54d8081bc32f2742a9b3c738f0821191  models/mel-roformer-zfturbo-vocals-v1/model.safetensors
+d8df573c2b0248e3ce77eb8f0fc3c528c9db5368f38335e249e25e05e84415ba  models/uvr-mdx-net-inst-hq3/config.json
+72c307239fbc415c9f3c1ef83fee38d19ac98b4b893f1a5afc4345296f48eea8  models/uvr-mdx-net-inst-hq3/model.safetensors
+8e4ff22dba85aa83e92ac8efa1f4d741c9e0cc91c9d4c51ca631c91177f80109  validation/automatic-downloads.json
+ee237e5eb45f11ee6684cdaec3e0deb20bee06ca5025e1af8be73d84756536c2  validation/bs-roformer-sw.json
+b10e9d5117578b4e69220ffeb1ccd339fed595411014f30f8697af2dc0ff86d1  validation/htdemucs.json
+b1be8b6a2d2c03218fbb9b58c9608f471f04b3fefd7d11097eb12a33f3eb38cc  validation/htdemucs_6s.json
+1b766b8401bb198ec8aec3e963a87b0dbe7536326cf3f3855e2c6d38d24497ea  validation/htdemucs_ft.json
+e4cd78ec803f05b27f02333d953d44f8f55e8411b9e40b2ec37249bec12be071  validation/mel-roformer-instv7n.json
+aeab6734ec12ba55abed0be71fe9d812970585a03cc37b1f35dd8ae1970cf0bd  validation/mel-roformer-zfturbo-vocals-v1.json
+8a552063129fa354dd4c1622de93962ec104e97e7b1fc768b56f1984aacd24d2  validation/uvr-mdx-net-inst-hq3.json

THIRD_PARTY_NOTICES.md

@@ -0,0 +1,138 @@
+# Third-Party Notices
+
+This repository distributes pretrained music-separation weights, their configurations,
+model cards, validation reports, and a catalog helper. It brings together work by the
+model authors, upstream projects, and MLX conversion contributors identified below.
+
+Licensing information is recorded per model. [LICENSE-code](LICENSE-code) covers the
+catalog helper; it does not apply a single license to all checkpoint weights. Existing
+model-specific notices are included with the corresponding files. Source revisions,
+filenames, and SHA-256 hashes are recorded in [catalog.json](catalog.json).
+
+## BS-RoFormer-SW
+
+- **Catalog entry:** [BS-RoFormer-SW](cards/bs-roformer-sw.md).
+- **Fixed checkpoint lineage:** jarredou's distribution, mirrored by
+  [enerjazzer/BS-ROFO-SW-Fixed](https://huggingface.co/enerjazzer/BS-ROFO-SW-Fixed/tree/a443a2985534b3bc815ef54a5d446c6a0390f974).
+  These names identify the checkpoint's distribution history, not a verified training-author credit.
+- **Conversion input:** `BS-Roformer-SW.ckpt` from the
+  [python-audio-separator model-configs release](https://github.com/nomadkaraoke/python-audio-separator/releases/download/model-configs/BS-Roformer-SW.ckpt).
+  Its SHA-256 matches the pinned fixed checkpoint above.
+- **MLX conversion:** ssmall256; float32 safetensors and the matching configuration
+  prepared with [mlx-audio-separator](https://github.com/ssmall256/mlx-audio-separator).
+- **Weight license information:** the pinned source model card explicitly declares
+  `license: unknown`. This catalog does not assign an MIT license to these weights.
+
+## Mel-RoFormer InstV7n
+
+- **Catalog entry:** [Mel-RoFormer InstV7n](cards/mel-roformer-instv7n.md).
+- **Checkpoint author:** Gabox. The matching checkpoint is available as
+  [`INSTV7N.ckpt`](https://huggingface.co/GaboxR67/MelBandRoformers/blob/7860d31d8cc1c73b46c1cbdc2b497210c97478ae/melbandroformers/instrumental/INSTV7N.ckpt)
+  in Gabox's model repository.
+- **Conversion input:** `mel_band_roformer_instrumental_instv7n_gabox.ckpt` and its
+  configuration from the
+  [python-audio-separator model-configs release](https://github.com/nomadkaraoke/python-audio-separator/releases/tag/model-configs).
+- **MLX conversion:** ssmall256; float32 safetensors and the matching configuration
+  prepared with mlx-audio-separator.
+- **Weight license information:** no weight-specific redistribution license is recorded
+  in the source metadata used for this catalog.
+
+## UVR-MDX-NET Inst HQ3
+
+- **Catalog entry:** [UVR-MDX-NET Inst HQ3](cards/uvr-mdx-net-inst-hq3.md).
+- **Checkpoint source:** the Ultimate Vocal Remover community and
+  [TRvlvr's public UVR model release](https://github.com/TRvlvr/model_repo/releases/tag/all_public_uvr_models).
+- **Conversion input:** `UVR-MDX-NET-Inst_HQ_3.onnx`, with model parameters from the
+  upstream UVR catalog used by python-audio-separator.
+- **MLX conversion:** ssmall256; float32 safetensors and the matching configuration
+  prepared with mlx-audio-separator.
+- **Weight license information:** no weight-specific redistribution license is recorded
+  in the source metadata used for this catalog.
+
+## HTDemucs, HTDemucs_FT, and HTDemucs_6s
+
+- **Catalog entries:** [HTDemucs](cards/htdemucs.md), [HTDemucs_FT](cards/htdemucs_ft.md),
+  and [HTDemucs_6s](cards/htdemucs_6s.md).
+- **Original project and models:** Meta Platforms, Inc. and affiliates, Alexandre
+  Défossez, and the [Demucs contributors](https://github.com/facebookresearch/demucs).
+- **MLX conversion and packaging:** ssmall256,
+  [demucs-mlx](https://github.com/ssmall256/demucs-mlx).
+- **Distributed artifacts:** unchanged float32 weights and configurations from
+  [ssmall256/demucs-mlx at revision `35ecb1a`](https://huggingface.co/ssmall256/demucs-mlx/tree/35ecb1a906ec420b1a28069db575ad070d35860e).
+  HTDemucs_FT includes all four specialist models.
+- **License information:** the pinned source repository declares MIT. Its copyright
+  and license text is preserved unchanged in each model directory:
+  [HTDemucs LICENSE](models/htdemucs/LICENSE),
+  [HTDemucs_FT LICENSE](models/htdemucs_ft/LICENSE), and
+  [HTDemucs_6s LICENSE](models/htdemucs_6s/LICENSE).
+
+The accompanying notices credit:
+
+> Copyright (c) Meta Platforms, Inc. and affiliates.
+> Copyright (c) 2026 ssmall256
+
+## ZFTurbo Mel-RoFormer Vocals v1
+
+- **Catalog entry:** [ZFTurbo Mel-RoFormer Vocals v1](cards/mel-roformer-zfturbo-vocals-v1.md).
+- **Original checkpoint:** Roman Solovyev (ZFTurbo),
+  [`model_vocals_mel_band_roformer_sdr_8.42.ckpt`](https://github.com/ZFTurbo/Music-Source-Separation-Training/releases/tag/v1.0.0)
+  from Music-Source-Separation-Training v1.0.0.
+- **MLX conversion and packaging:** Xocialize, published through
+  [mlx-community/mel-roformer-zfturbo-vocals-v1-mlx](https://huggingface.co/mlx-community/mel-roformer-zfturbo-vocals-v1-mlx/tree/01bef2daebedf39937f552d6da68340fc979c9a0).
+- **Distributed artifacts:** the pinned float16 weights and configuration are copied
+  unchanged. This catalog does not reconvert or quantize them.
+- **License information:** the pinned model card declares MIT. The included
+  [LICENSE](models/mel-roformer-zfturbo-vocals-v1/LICENSE) preserves its copyright
+  notices and disclosure: the source repository's MIT license was added on
+  November 4, 2024, after the November 6, 2023 checkpoint release. The source
+  redistribution treats the weights as inheriting the repository's current MIT terms;
+  explicit retroactive author confirmation was not collected for this catalog.
+
+## Architecture and implementation credits
+
+The BS-RoFormer, Mel-Band-RoFormer, Demucs, and MDX/UVR architecture authors and
+contributors retain credit for their work. Architecture names describe the model
+families; they do not establish authorship or licensing of every trained checkpoint.
+
+[python-audio-separator](https://github.com/nomadkaraoke/python-audio-separator),
+maintained by Nomad Karaoke / karaokenerds and contributors, is the upstream project
+for mlx-audio-separator. The MLX application adapts its model/configuration handling
+conventions and uses it for PyTorch reference inference. Its model-configs release
+also supplies the SW and InstV7n conversion inputs identified above. The application
+source is maintained separately on GitHub.
+
+The upstream notice is reproduced below as attribution for that software. It is not
+presented as a license grant for third-party checkpoints distributed through its catalog.
+
+### MIT License — python-audio-separator
+
+Copyright (c) 2023 karaokenerds
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+## Using and redistributing these files
+
+Keep the matching model configuration and applicable notices with each set of weights.
+Refer to the model-specific sources above for terms relevant to your use. An unknown
+or undocumented weight license is not a grant of permission, and the helper's code
+license does not fill that gap.
+
+For attribution corrections, please open an issue in
+[mlx-audio-separator](https://github.com/ssmall256/mlx-audio-separator/issues), identifying
+the catalog model and the relevant original source.

VALIDATION.md

@@ -0,0 +1,84 @@
+# Validation
+
+The catalog includes an individual JSON report for each of its seven models, linked below. The reports
+record input and output shapes, inference settings, output peaks, file-writing results, and
+conversion comparisons where available.
+
+## Catalog checks
+
+All seven models passed a fresh-process inference check on the same 14-second stereo music
+excerpt at 44.1 kHz, with overlapping chunks. Each produced the expected number of finite,
+full-length stereo stems and wrote its output audio successfully. Network access was disabled,
+and PyTorch, ONNX, and ONNX Runtime imports were blocked during inference.
+
+| Model | Stems checked | Additional verification | Report |
+|---|---:|---|---|
+| BS-RoFormer-SW | 6 | Bit-identical unquantized output versus the original checkpoint in MLX | [JSON](validation/bs-roformer-sw.json) |
+| Mel-RoFormer InstV7n | 2 | Bit-identical unquantized output versus the original checkpoint in MLX | [JSON](validation/mel-roformer-instv7n.json) |
+| UVR-MDX-NET Inst HQ3 | 2 | Bit-identical unquantized output versus the original checkpoint in MLX | [JSON](validation/uvr-mdx-net-inst-hq3.json) |
+| HTDemucs | 4 | Weights and config match the pinned published files | [JSON](validation/htdemucs.json) |
+| HTDemucs_FT | 4 | Weights and config match the pinned published files | [JSON](validation/htdemucs_ft.json) |
+| HTDemucs_6s | 6 | Weights and config match the pinned published files | [JSON](validation/htdemucs_6s.json) |
+| ZFTurbo Vocals v1 | 2 | Weights and config match the pinned published files | [JSON](validation/mel-roformer-zfturbo-vocals-v1.json) |
+
+The three new conversions were compared before output quantization, across all ten emitted
+stems. The four reused models were downloaded into an empty cache and verified by size and
+SHA-256 before inference. Their weights retain the original precision: float32 for Demucs,
+float16 for ZFTurbo.
+
+RoFormer checks used two overlapping windows. Demucs checks used two shifts, a fixed seed of
+zero, and 25% overlap. The fixed seed makes the validation runs repeatable; normal package use leaves shift selection random unless a seed is supplied. The source audio and separated
+music are not distributed with the models.
+
+## Automatic downloads and offline loading
+
+The [public API integration report](validation/automatic-downloads.json) covers all seven models
+through normal `Separator.load_model()` and `Separator.separate()` calls. Each test:
+
+1. Starts with an empty model directory and downloads the exact repository files from a local HTTP mirror.
+2. Loads the model using its existing package name and separates the 14-second fixture.
+3. Blocks all network connections, creates a new separator, and loads the same cache using the short catalog ID.
+4. Verifies full-length, finite output and bit-identical unquantized stems between both runs.
+
+PyTorch, ONNX, and ONNX Runtime imports are blocked throughout. The test checks actual audio writes
+and confirms that each model file is requested exactly once. This exercises download and cache
+behavior independently of Hugging Face availability. The report records the tested implementation's
+source hashes and identifies the local mirror as the download source.
+
+CPU tests additionally cover corrupt downloads and caches, interrupted transfers, concurrent
+first use, offline cache misses, older cache reuse, and unchanged routing for other models.
+
+## Earlier reference comparisons
+
+The following evidence comes from the separator projects' reference tests, independently of
+the catalog packaging checks. PyTorch references use the original
+[python-audio-separator](https://github.com/nomadkaraoke/python-audio-separator).
+
+- **BS-RoFormer-SW:** the standard core passes real-checkpoint tests against native
+  python-audio-separator references and cross-repository consistency checks. Earlier full-track
+  PCM comparisons include near-silent relative-error exceptions bounded by one PCM16 step.
+- **InstV7n and Inst HQ3:** a twelve-track comparison on two hosts passed relative L2 ≤ 0.05
+  for every compared stem. The worst values were 0.0120522 and 0.0047207, respectively.
+- **HTDemucs and HTDemucs_FT:** full-track controls with identical shift offsets, following the
+  normalization correction, had worst relative L2 errors of 0.000775014 and 0.00106586.
+  These are single-track controls. Other corpus comparisons include stochastic shift differences.
+- **HTDemucs_6s:** the existing Demucs model publication reports source-checkpoint parity.
+  The catalog check adds offline six-stem inference and exact file verification; it does not
+  rerun the PyTorch reference comparison.
+- **ZFTurbo Vocals v1:** earlier release tests cover both stems, overlap, input sample rates,
+  and mixture reconstruction. This catalog uses the same published float16 conversion.
+
+Detailed methods and results are in the mlx-audio-separator
+[performance report](https://github.com/ssmall256/mlx-audio-separator/blob/main/docs/performance-snapshot.md),
+[release validation](https://github.com/ssmall256/mlx-audio-separator/blob/main/docs/release-validation.md),
+and [RoFormer core documentation](https://github.com/ssmall256/mlx-audio-separator/blob/main/docs/roformer-core.md).
+
+## Interpreting the results
+
+The catalog checks verify that the packaged files load and preserve the tested MLX behavior.
+They do not provide a new ground-truth SDR score, guarantee separation quality on every song,
+or establish a speedup from the recorded validation wall times. Bit-identical conversion
+results describe original-MLX versus safetensors-MLX outputs, not MLX versus PyTorch.
+
+The [manifest](catalog.json) records source identities, runtime versions, and conversion source
+fingerprints. The model cards list file hashes, and `SHA256SUMS` covers the distributed files.

cards/bs-roformer-sw.md

@@ -0,0 +1,59 @@
+# BS-RoFormer-SW
+
+A six-stem BS-RoFormer model for separating bass, drums, other instruments, vocals, guitar, and piano. These float32 safetensors run directly in MLX on Apple Silicon.
+
+| Property | Value |
+|---|---|
+| Catalog ID | `bs-roformer-sw` |
+| Existing package name | `BS-Roformer-SW.ckpt` |
+| Output stems | bass, drums, other, vocals, guitar, piano |
+| Output audio | Stereo, 44,100 Hz |
+| Stored weight precision | `float32` |
+| Download size | 666.5 MiB |
+
+## Use this model
+
+Install [mlx-audio-separator](../README.md#quick-start), then run:
+
+```bash
+mlx-audio-separator song.m4a -m bs-roformer-sw --output_dir stems
+```
+
+The first use automatically downloads and verifies this model's safetensors and configuration.
+Later runs reuse the local cache. The existing name `BS-Roformer-SW.ckpt` selects the same model.
+For Python, use `separator.load_model("bs-roformer-sw")`.
+See [direct downloads](../README.md#direct-downloads) to prepare the files manually or use them offline.
+
+Chunks contain 588,800 samples (about 13.35 seconds). Two overlapping windows give a half-chunk stride. Inference runs locally on Apple Silicon through `mlx-audio-separator`.
+
+## Files and provenance
+
+Keep the weights and configuration together in `models/bs-roformer-sw/`.
+
+| File | Bytes | SHA-256 |
+|---|---:|---|
+| [`model.safetensors`](../models/bs-roformer-sw/model.safetensors) | 698,841,589 | `f95907cd5bfdbc783c1cb32f6aafc49b8b6a598477a9d4a48b4f9191c5167707` |
+| [`config.json`](../models/bs-roformer-sw/config.json) | 4,282 | `386a86c2061536d2dca349943f1027f3ac5472ccb22b0e18bd528519b90f8931` |
+
+The [catalog manifest](../catalog.json) records source hashes, conversion details, and immutable
+upstream revisions where applicable. [Original source artifacts](https://github.com/nomadkaraoke/python-audio-separator/releases/download/model-configs/BS-Roformer-SW.ckpt).
+
+## Validation
+
+The [inference report](../validation/bs-roformer-sw.json) records 6 finite,
+full-length stereo stems from a 14-second real-music excerpt with overlapping chunks.
+Network access and PyTorch, ONNX, and ONNX Runtime imports were disabled during the check.
+
+All unquantized output samples were **bit-identical** to the original checkpoint loaded through the same MLX implementation.
+
+The normal package API also passed first-use download and offline inference checks with
+bit-identical output; see the [automatic-download report](../validation/automatic-downloads.json).
+
+See [Validation details](../VALIDATION.md) for the scope of these checks and earlier reference
+comparisons. This packaging check does not provide a new separation-quality score or speed benchmark.
+
+## License and attribution
+
+The source catalog credits **jarredou**. **enerjazzer** hosts the fixed source checkpoint. The [source model card](https://huggingface.co/enerjazzer/BS-ROFO-SW-Fixed/tree/a443a2985534b3bc815ef54a5d446c6a0390f974) explicitly lists the weight license as **unknown**. The catalog helper's MIT license does not establish a license for these weights.
+
+See the [collection's licensing and credits](../README.md#licensing-and-credit) for additional sources.

cards/htdemucs.md

@@ -0,0 +1,60 @@
+# HTDemucs
+
+The four-stem Hybrid Transformer Demucs model: drums, bass, other instruments, and vocals. This entry contains the float32 MLX conversion from ssmall256/demucs-mlx.
+
+| Property | Value |
+|---|---|
+| Catalog ID | `htdemucs` |
+| Existing package name | `htdemucs.yaml` |
+| Output stems | drums, bass, other, vocals |
+| Output audio | Stereo, 44,100 Hz |
+| Stored weight precision | `float32` |
+| Download size | 160.2 MiB |
+
+## Use this model
+
+Install [mlx-audio-separator](../README.md#quick-start), then run:
+
+```bash
+mlx-audio-separator song.m4a -m htdemucs --output_dir stems
+```
+
+The first use automatically downloads and verifies this model's safetensors and configuration.
+Later runs reuse the local cache. The existing name `htdemucs.yaml` selects the same model.
+For Python, use `separator.load_model("htdemucs")`.
+See [direct downloads](../README.md#direct-downloads) to prepare the files manually or use them offline.
+
+The package uses two shifts and 25% overlap. Set `--demucs_seed 0` for repeatable shift selection. Inference runs locally on Apple Silicon through `mlx-audio-separator`.
+
+## Files and provenance
+
+Keep the weights and configuration together in `models/htdemucs/`.
+
+| File | Bytes | SHA-256 |
+|---|---:|---|
+| [`htdemucs.safetensors`](../models/htdemucs/htdemucs.safetensors) | 168,005,865 | `339d267a7a6983a11eedbdc00413c602a65e9b9103f695fb5c2b2a481cd9d297` |
+| [`htdemucs_config.json`](../models/htdemucs/htdemucs_config.json) | 4,215 | `23657b19db14771aecf366ceedbf846c1386836c10043c97f4569acf05526b78` |
+| [`LICENSE`](../models/htdemucs/LICENSE) | 1,117 | `15086279d32c0f00c577c0f52ff428daf98b8a1fec0264da1c717c88ad464f51` |
+
+The [catalog manifest](../catalog.json) records source hashes, conversion details, and immutable
+upstream revisions where applicable. [Original source artifacts](https://huggingface.co/ssmall256/demucs-mlx/tree/35ecb1a906ec420b1a28069db575ad070d35860e).
+
+## Validation
+
+The [inference report](../validation/htdemucs.json) records 4 finite,
+full-length stereo stems from a 14-second real-music excerpt with overlapping chunks.
+Network access and PyTorch, ONNX, and ONNX Runtime imports were disabled during the check.
+
+Weights and configuration match the pinned published files by SHA-256. The original precision and file contents are unchanged.
+
+The normal package API also passed first-use download and offline inference checks with
+bit-identical output; see the [automatic-download report](../validation/automatic-downloads.json).
+
+See [Validation details](../VALIDATION.md) for the scope of these checks and earlier reference
+comparisons. This packaging check does not provide a new separation-quality score or speed benchmark.
+
+## License and attribution
+
+**Meta and the Demucs contributors** developed the original model. The pinned [Demucs MLX model repository](https://huggingface.co/ssmall256/demucs-mlx/tree/35ecb1a906ec420b1a28069db575ad070d35860e) declares **MIT**. The accompanying [license notice](../models/htdemucs/LICENSE) is included unchanged.
+
+See the [collection's licensing and credits](../README.md#licensing-and-credit) for additional sources.

cards/htdemucs_6s.md

@@ -0,0 +1,60 @@
+# HTDemucs_6s
+
+The six-stem Hybrid Transformer Demucs model adds guitar and piano to drums, bass, other instruments, and vocals. This entry contains the float32 MLX conversion from ssmall256/demucs-mlx.
+
+| Property | Value |
+|---|---|
+| Catalog ID | `htdemucs_6s` |
+| Existing package name | `htdemucs_6s.yaml` |
+| Output stems | drums, bass, other, vocals, guitar, piano |
+| Output audio | Stereo, 44,100 Hz |
+| Stored weight precision | `float32` |
+| Download size | 104.6 MiB |
+
+## Use this model
+
+Install [mlx-audio-separator](../README.md#quick-start), then run:
+
+```bash
+mlx-audio-separator song.m4a -m htdemucs_6s --output_dir stems
+```
+
+The first use automatically downloads and verifies this model's safetensors and configuration.
+Later runs reuse the local cache. The existing name `htdemucs_6s.yaml` selects the same model.
+For Python, use `separator.load_model("htdemucs_6s")`.
+See [direct downloads](../README.md#direct-downloads) to prepare the files manually or use them offline.
+
+The package uses two shifts and 25% overlap. Set `--demucs_seed 0` for repeatable shift selection. Inference runs locally on Apple Silicon through `mlx-audio-separator`.
+
+## Files and provenance
+
+Keep the weights and configuration together in `models/htdemucs_6s/`.
+
+| File | Bytes | SHA-256 |
+|---|---:|---|
+| [`htdemucs_6s.safetensors`](../models/htdemucs_6s/htdemucs_6s.safetensors) | 109,726,583 | `d298f7f746bf53c21baad44fb08e88807ef47feb551dd22f1601a546c85b8e02` |
+| [`htdemucs_6s_config.json`](../models/htdemucs_6s/htdemucs_6s_config.json) | 4,302 | `d5e18c0209be583027d6eb5ec3013a02cd7ef55bc3096e644d8d3577d92e9bd0` |
+| [`LICENSE`](../models/htdemucs_6s/LICENSE) | 1,117 | `15086279d32c0f00c577c0f52ff428daf98b8a1fec0264da1c717c88ad464f51` |
+
+The [catalog manifest](../catalog.json) records source hashes, conversion details, and immutable
+upstream revisions where applicable. [Original source artifacts](https://huggingface.co/ssmall256/demucs-mlx/tree/35ecb1a906ec420b1a28069db575ad070d35860e).
+
+## Validation
+
+The [inference report](../validation/htdemucs_6s.json) records 6 finite,
+full-length stereo stems from a 14-second real-music excerpt with overlapping chunks.
+Network access and PyTorch, ONNX, and ONNX Runtime imports were disabled during the check.
+
+Weights and configuration match the pinned published files by SHA-256. The original precision and file contents are unchanged.
+
+The normal package API also passed first-use download and offline inference checks with
+bit-identical output; see the [automatic-download report](../validation/automatic-downloads.json).
+
+See [Validation details](../VALIDATION.md) for the scope of these checks and earlier reference
+comparisons. This packaging check does not provide a new separation-quality score or speed benchmark.
+
+## License and attribution
+
+**Meta and the Demucs contributors** developed the original model. The pinned [Demucs MLX model repository](https://huggingface.co/ssmall256/demucs-mlx/tree/35ecb1a906ec420b1a28069db575ad070d35860e) declares **MIT**. The accompanying [license notice](../models/htdemucs_6s/LICENSE) is included unchanged.
+
+See the [collection's licensing and credits](../README.md#licensing-and-credit) for additional sources.

cards/htdemucs_ft.md

@@ -0,0 +1,60 @@
+# HTDemucs_FT
+
+The fine-tuned four-stem Hybrid Transformer Demucs ensemble. It uses a specialist model for each of drums, bass, other instruments, and vocals. This entry contains the float32 MLX conversion from ssmall256/demucs-mlx.
+
+| Property | Value |
+|---|---|
+| Catalog ID | `htdemucs_ft` |
+| Existing package name | `htdemucs_ft.yaml` |
+| Output stems | drums, bass, other, vocals |
+| Output audio | Stereo, 44,100 Hz |
+| Stored weight precision | `float32` |
+| Download size | 640.9 MiB |
+
+## Use this model
+
+Install [mlx-audio-separator](../README.md#quick-start), then run:
+
+```bash
+mlx-audio-separator song.m4a -m htdemucs_ft --output_dir stems
+```
+
+The first use automatically downloads and verifies this model's safetensors and configuration.
+Later runs reuse the local cache. The existing name `htdemucs_ft.yaml` selects the same model.
+For Python, use `separator.load_model("htdemucs_ft")`.
+See [direct downloads](../README.md#direct-downloads) to prepare the files manually or use them offline.
+
+The package runs all four specialists with two shifts and 25% overlap. Set `--demucs_seed 0` for repeatable shift selection. Inference runs locally on Apple Silicon through `mlx-audio-separator`.
+
+## Files and provenance
+
+Keep the weights and configuration together in `models/htdemucs_ft/`.
+
+| File | Bytes | SHA-256 |
+|---|---:|---|
+| [`htdemucs_ft.safetensors`](../models/htdemucs_ft/htdemucs_ft.safetensors) | 672,024,519 | `53f03b1ad4b4d211025a35da65460ba61a17547adf9c0544cad0ebcc8d7bbabb` |
+| [`htdemucs_ft_config.json`](../models/htdemucs_ft/htdemucs_ft_config.json) | 10,253 | `6708535a698d4d4dd285a6f0a80d9d3c228fdd6514e53bdf1ddbc2ee9b0fa872` |
+| [`LICENSE`](../models/htdemucs_ft/LICENSE) | 1,117 | `15086279d32c0f00c577c0f52ff428daf98b8a1fec0264da1c717c88ad464f51` |
+
+The [catalog manifest](../catalog.json) records source hashes, conversion details, and immutable
+upstream revisions where applicable. [Original source artifacts](https://huggingface.co/ssmall256/demucs-mlx/tree/35ecb1a906ec420b1a28069db575ad070d35860e).
+
+## Validation
+
+The [inference report](../validation/htdemucs_ft.json) records 4 finite,
+full-length stereo stems from a 14-second real-music excerpt with overlapping chunks.
+Network access and PyTorch, ONNX, and ONNX Runtime imports were disabled during the check.
+
+Weights and configuration match the pinned published files by SHA-256. The original precision and file contents are unchanged.
+
+The normal package API also passed first-use download and offline inference checks with
+bit-identical output; see the [automatic-download report](../validation/automatic-downloads.json).
+
+See [Validation details](../VALIDATION.md) for the scope of these checks and earlier reference
+comparisons. This packaging check does not provide a new separation-quality score or speed benchmark.
+
+## License and attribution
+
+**Meta and the Demucs contributors** developed the original models. The pinned [Demucs MLX model repository](https://huggingface.co/ssmall256/demucs-mlx/tree/35ecb1a906ec420b1a28069db575ad070d35860e) declares **MIT**. The accompanying [license notice](../models/htdemucs_ft/LICENSE) is included unchanged.
+
+See the [collection's licensing and credits](../README.md#licensing-and-credit) for additional sources.

cards/mel-roformer-instv7n.md

@@ -0,0 +1,59 @@
+# Mel-RoFormer InstV7n (Gabox)
+
+An instrumental and vocal separation model by Gabox, converted to float32 safetensors for MLX inference on Apple Silicon.
+
+| Property | Value |
+|---|---|
+| Catalog ID | `mel-roformer-instv7n` |
+| Existing package name | `mel_band_roformer_instrumental_instv7n_gabox.ckpt` |
+| Output stems | Instrumental, Vocals |
+| Output audio | Stereo, 44,100 Hz |
+| Stored weight precision | `float32` |
+| Download size | 870.6 MiB |
+
+## Use this model
+
+Install [mlx-audio-separator](../README.md#quick-start), then run:
+
+```bash
+mlx-audio-separator song.m4a -m mel-roformer-instv7n --output_dir stems
+```
+
+The first use automatically downloads and verifies this model's safetensors and configuration.
+Later runs reuse the local cache. The existing name `mel_band_roformer_instrumental_instv7n_gabox.ckpt` selects the same model.
+For Python, use `separator.load_model("mel-roformer-instv7n")`.
+See [direct downloads](../README.md#direct-downloads) to prepare the files manually or use them offline.
+
+Chunks contain 485,100 samples (11 seconds). Two overlapping windows give a half-chunk stride. Inference runs locally on Apple Silicon through `mlx-audio-separator`.
+
+## Files and provenance
+
+Keep the weights and configuration together in `models/mel-roformer-instv7n/`.
+
+| File | Bytes | SHA-256 |
+|---|---:|---|
+| [`model.safetensors`](../models/mel-roformer-instv7n/model.safetensors) | 912,889,223 | `74397de5395ed372059b971d6398106f60d92373ada69b62e986e52c9697a5ba` |
+| [`config.json`](../models/mel-roformer-instv7n/config.json) | 1,150 | `cd8548fd6a1345cfd0c51970919e15e7262cee954b6ff45bb016ed1364be38c5` |
+
+The [catalog manifest](../catalog.json) records source hashes, conversion details, and immutable
+upstream revisions where applicable. [Original source artifacts](https://github.com/nomadkaraoke/python-audio-separator/releases/download/model-configs/mel_band_roformer_instrumental_instv7n_gabox.ckpt).
+
+## Validation
+
+The [inference report](../validation/mel-roformer-instv7n.json) records 2 finite,
+full-length stereo stems from a 14-second real-music excerpt with overlapping chunks.
+Network access and PyTorch, ONNX, and ONNX Runtime imports were disabled during the check.
+
+All unquantized output samples were **bit-identical** to the original checkpoint loaded through the same MLX implementation.
+
+The normal package API also passed first-use download and offline inference checks with
+bit-identical output; see the [automatic-download report](../validation/automatic-downloads.json).
+
+See [Validation details](../VALIDATION.md) for the scope of these checks and earlier reference
+comparisons. This packaging check does not provide a new separation-quality score or speed benchmark.
+
+## License and attribution
+
+The checkpoint is credited to **Gabox** and distributed through the upstream **python-audio-separator** model-configs release. No weight-specific redistribution license is recorded in the source metadata. The catalog helper's MIT license applies to its code only.
+
+See the [collection's licensing and credits](../README.md#licensing-and-credit) for additional sources.

cards/mel-roformer-zfturbo-vocals-v1.md

@@ -0,0 +1,60 @@
+# ZFTurbo Mel-RoFormer Vocals v1
+
+A vocal separation model by Roman Solovyev (ZFTurbo). The separator predicts vocals and derives the instrumental from the mixture. This entry preserves the published mlx-community float16 conversion.
+
+| Property | Value |
+|---|---|
+| Catalog ID | `mel-roformer-zfturbo-vocals-v1` |
+| Existing package name | `mel-roformer-zfturbo-vocals-v1-mlx` |
+| Output stems | Vocals, Instrumental |
+| Output audio | Stereo, 44,100 Hz |
+| Stored weight precision | `float16` |
+| Download size | 64.3 MiB |
+
+## Use this model
+
+Install [mlx-audio-separator](../README.md#quick-start), then run:
+
+```bash
+mlx-audio-separator song.m4a -m mel-roformer-zfturbo-vocals-v1 --output_dir stems
+```
+
+The first use automatically downloads and verifies this model's safetensors and configuration.
+Later runs reuse the local cache. The existing name `mel-roformer-zfturbo-vocals-v1-mlx` selects the same model.
+For Python, use `separator.load_model("mel-roformer-zfturbo-vocals-v1")`.
+See [direct downloads](../README.md#direct-downloads) to prepare the files manually or use them offline.
+
+Chunks contain 352,800 samples (8 seconds). Two overlapping windows give a half-chunk stride. Output is always 44.1 kHz. Inference runs locally on Apple Silicon through `mlx-audio-separator`.
+
+## Files and provenance
+
+Keep the weights and configuration together in `models/mel-roformer-zfturbo-vocals-v1/`.
+
+| File | Bytes | SHA-256 |
+|---|---:|---|
+| [`model.safetensors`](../models/mel-roformer-zfturbo-vocals-v1/model.safetensors) | 67,402,202 | `ef4aa052845a868cfaff93611477bd8f54d8081bc32f2742a9b3c738f0821191` |
+| [`config.json`](../models/mel-roformer-zfturbo-vocals-v1/config.json) | 1,016 | `d110ed4713b403d27e2ae6c7796b1a864576232de4a2070e9762108ca197caa6` |
+| [`LICENSE`](../models/mel-roformer-zfturbo-vocals-v1/LICENSE) | 1,779 | `2b55ae02dcfe9ac8454c41c5e235175512db7429ee7275bab4c4e566b3f80597` |
+
+The [catalog manifest](../catalog.json) records source hashes, conversion details, and immutable
+upstream revisions where applicable. [Original source artifacts](https://huggingface.co/mlx-community/mel-roformer-zfturbo-vocals-v1-mlx/tree/01bef2daebedf39937f552d6da68340fc979c9a0).
+
+## Validation
+
+The [inference report](../validation/mel-roformer-zfturbo-vocals-v1.json) records 2 finite,
+full-length stereo stems from a 14-second real-music excerpt with overlapping chunks.
+Network access and PyTorch, ONNX, and ONNX Runtime imports were disabled during the check.
+
+Weights and configuration match the pinned published files by SHA-256. The original precision and file contents are unchanged.
+
+The normal package API also passed first-use download and offline inference checks with
+bit-identical output; see the [automatic-download report](../validation/automatic-downloads.json).
+
+See [Validation details](../VALIDATION.md) for the scope of these checks and earlier reference
+comparisons. This packaging check does not provide a new separation-quality score or speed benchmark.
+
+## License and attribution
+
+The original checkpoint is by **Roman Solovyev (ZFTurbo)**; the existing MLX conversion is from **mlx-community**. Its [model card](https://huggingface.co/mlx-community/mel-roformer-zfturbo-vocals-v1-mlx/tree/01bef2daebedf39937f552d6da68340fc979c9a0) declares **MIT**, noting that the source license was added after the checkpoint release and that explicit retroactive author confirmation was not collected. The [license notice](../models/mel-roformer-zfturbo-vocals-v1/LICENSE) preserves that caveat.
+
+See the [collection's licensing and credits](../README.md#licensing-and-credit) for additional sources.

cards/uvr-mdx-net-inst-hq3.md

@@ -0,0 +1,59 @@
+# UVR-MDX-NET Inst HQ3
+
+An instrumental and vocal separation model from the UVR catalog. This float32 safetensors conversion runs in MLX and does not require ONNX Runtime.
+
+| Property | Value |
+|---|---|
+| Catalog ID | `uvr-mdx-net-inst-hq3` |
+| Existing package name | `UVR-MDX-NET-Inst_HQ_3.onnx` |
+| Output stems | Instrumental, Vocals |
+| Output audio | Stereo, 44,100 Hz |
+| Stored weight precision | `float32` |
+| Download size | 63.8 MiB |
+
+## Use this model
+
+Install [mlx-audio-separator](../README.md#quick-start), then run:
+
+```bash
+mlx-audio-separator song.m4a -m uvr-mdx-net-inst-hq3 --output_dir stems
+```
+
+The first use automatically downloads and verifies this model's safetensors and configuration.
+Later runs reuse the local cache. The existing name `UVR-MDX-NET-Inst_HQ_3.onnx` selects the same model.
+For Python, use `separator.load_model("uvr-mdx-net-inst-hq3")`.
+See [direct downloads](../README.md#direct-downloads) to prepare the files manually or use them offline.
+
+The package uses 25% overlap and a segment size of 256. Inference runs locally on Apple Silicon through `mlx-audio-separator`.
+
+## Files and provenance
+
+Keep the weights and configuration together in `models/uvr-mdx-net-inst-hq3/`.
+
+| File | Bytes | SHA-256 |
+|---|---:|---|
+| [`model.safetensors`](../models/uvr-mdx-net-inst-hq3/model.safetensors) | 66,871,394 | `72c307239fbc415c9f3c1ef83fee38d19ac98b4b893f1a5afc4345296f48eea8` |
+| [`config.json`](../models/uvr-mdx-net-inst-hq3/config.json) | 204 | `d8df573c2b0248e3ce77eb8f0fc3c528c9db5368f38335e249e25e05e84415ba` |
+
+The [catalog manifest](../catalog.json) records source hashes, conversion details, and immutable
+upstream revisions where applicable. [Original source artifacts](https://github.com/TRvlvr/model_repo/releases/download/all_public_uvr_models/UVR-MDX-NET-Inst_HQ_3.onnx).
+
+## Validation
+
+The [inference report](../validation/uvr-mdx-net-inst-hq3.json) records 2 finite,
+full-length stereo stems from a 14-second real-music excerpt with overlapping chunks.
+Network access and PyTorch, ONNX, and ONNX Runtime imports were disabled during the check.
+
+All unquantized output samples were **bit-identical** to the original checkpoint loaded through the same MLX implementation.
+
+The normal package API also passed first-use download and offline inference checks with
+bit-identical output; see the [automatic-download report](../validation/automatic-downloads.json).
+
+See [Validation details](../VALIDATION.md) for the scope of these checks and earlier reference
+comparisons. This packaging check does not provide a new separation-quality score or speed benchmark.
+
+## License and attribution
+
+The checkpoint comes from the **TRvlvr/UVR** public model release. No weight-specific redistribution license is recorded in the source metadata. The catalog helper's MIT license applies to its code only.
+
+See the [collection's licensing and credits](../README.md#licensing-and-credit) for additional sources.

catalog.json

@@ -0,0 +1,552 @@
+{
+  "catalog_version": "1-draft",
+  "conversion": {
+    "base_commit": "95a1effb864700a4156f1e199771b58c89da8848",
+    "repository": "https://github.com/ssmall256/mlx-audio-separator",
+    "source_sha256": {
+      "mlx_audio_separator/__init__.py": "7a60ef69c42996da1512c584a0fced4282d96b493aff9b7522d043f36556f9c0",
+      "mlx_audio_separator/__main__.py": "86f9909b5db882b7c1f51007adac5ebe62c24d22f224cc9d2fa521e4529cfbe5",
+      "mlx_audio_separator/core.py": "6b00478e07f69a5b3a35dfce683ebc42bdd2ac66fe6855557c48a7ddb56615ca",
+      "mlx_audio_separator/demucs_mlx/__init__.py": "1cb0541a734668ef204bd3a8bb9e4efeea0f32488b2e0ee0db114ee15941a232",
+      "mlx_audio_separator/demucs_mlx/api.py": "e1328d71c1d81fe5159504ede9f317dc8753230a7abffac391803588ae74fab9",
+      "mlx_audio_separator/demucs_mlx/apply_mlx.py": "43b9d6b674e53054a6190ff5958c63f9bbdd0e15d44d09646988bb4fa6dfed78",
+      "mlx_audio_separator/demucs_mlx/audio.py": "0a419eccc33ab537a9626179d7fc480954a38591c25330d26a6a9b0d2c9806a1",
+      "mlx_audio_separator/demucs_mlx/defaults.py": "207c6b62b8fe2927fb7e126f19899fca2cecb07bb7d68869eb0333cf03b7e030",
+      "mlx_audio_separator/demucs_mlx/hardware.py": "c8b91faa395528e9a98e7b35a394370fb47b609a1cd0433e2bc63974108ce69a",
+      "mlx_audio_separator/demucs_mlx/hub.py": "0ea01ab32275e5607fd47a8ce528b2ee7c9f7fec8d1e27ec294e3bab6dc37599",
+      "mlx_audio_separator/demucs_mlx/interop.py": "8c3b7c55e4b0117b9d830ea77f48979793d57d5c7c016b7fcf0036c4f9e232ab",
+      "mlx_audio_separator/demucs_mlx/io_pipeline.py": "c7c45d79cbde7d66f9f68d94a83855cf1e5fe36b256907ac36f325a4c27d463c",
+      "mlx_audio_separator/demucs_mlx/metal_kernels.py": "8dfb6f1db487b83752d4c2ec1d89f59241d7f0d28a247da4e8709208b02c8576",
+      "mlx_audio_separator/demucs_mlx/mlx_backend.py": "e4819443d771187f4b75f297b3b378568b93cc6b1c1b4cc1a568391e7d7bced0",
+      "mlx_audio_separator/demucs_mlx/mlx_convert.py": "c8a9d899cccc57306d27c9ae1de3c0f6fbdb624acbe855fcdfdc621a96ed1035",
+      "mlx_audio_separator/demucs_mlx/mlx_demucs.py": "0f6db07f2d8a1dff9bf4852b00135c8f06fe559a6e29a8ca3d40306d6c6bf4f0",
+      "mlx_audio_separator/demucs_mlx/mlx_hdemucs.py": "9371d5859977443b47a271e5b6e3550f21d022517a05cdae64d66922ce4d8f6e",
+      "mlx_audio_separator/demucs_mlx/mlx_htdemucs.py": "628d2aec74e8e6293d7b8983dcc84f8cafd2af37210b5ba86db738f59af7d683",
+      "mlx_audio_separator/demucs_mlx/mlx_layers.py": "f4ce3c9e28c41027e9885a48c71c3e573475861edc7d8f8f31a4ee3683284715",
+      "mlx_audio_separator/demucs_mlx/mlx_registry.py": "d50be9d6b64a238816302439ccbc95beb1c36d7c6afc277c3bbe83d607b72adc",
+      "mlx_audio_separator/demucs_mlx/mlx_transformer.py": "213918c3c82a40fd6bf0f45f5f3f4e4a99dcaf80009d1f4916ba713398f15495",
+      "mlx_audio_separator/demucs_mlx/mlx_utils.py": "100050b061783401ec83a90e98bf78e328365e9317ededbf41aeaaeb47f0e87e",
+      "mlx_audio_separator/demucs_mlx/model_converter.py": "ee1a65d777d96a74c50bfb7068dd21653467000d411de48d85e7310b0aba0753",
+      "mlx_audio_separator/demucs_mlx/secure_demucs.py": "289736dec3dc542b5f51d15ed12b287e8a555a96b06b2947d0a25816044d2f39",
+      "mlx_audio_separator/demucs_mlx/spec_mlx.py": "6de75340baec75263049c070d3eb4e2522207f588f885bc560462681ba4ff219",
+      "mlx_audio_separator/demucs_mlx/wiener_mlx.py": "11815ee06c9dfad0b55cb62bf4dc30c8b1e668c3dbe2e4a1b12a4318c5b85950",
+      "mlx_audio_separator/hf_mel_roformer.py": "0952448f4ef881ab51b1fa6ffa163b0cd3c298ab0d710e7b942e8671a8699a80",
+      "mlx_audio_separator/separator/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
+      "mlx_audio_separator/separator/architectures/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
+      "mlx_audio_separator/separator/architectures/demucs_separator.py": "4db7cb8b4367b132bcee6eaca224bbf066107e65191180413540f06cb6f0cccc",
+      "mlx_audio_separator/separator/architectures/mdx_separator.py": "80bcae233809e62526c5af7b7fdeb334bdf0df508fe60ef7a7ed025686929269",
+      "mlx_audio_separator/separator/architectures/mdxc_separator.py": "56a3cb6cb7e68a9038020d916e87d0106ae0bbc127cda690a2b99a76857f4758",
+      "mlx_audio_separator/separator/architectures/vr_separator.py": "bdd871e9751a753691cd760c283dcecb18d6ffd7ffad506c24709cb1499e7178",
+      "mlx_audio_separator/separator/audio_chunking.py": "cbe10541fbe43744753d0d387ee933d00daf370d89142df6c6d855f67aae56bc",
+      "mlx_audio_separator/separator/common_separator.py": "ddd1370848dd919b54dec495d98aa4f50f584b8978b51249334d4a0930824d8a",
+      "mlx_audio_separator/separator/models/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
+      "mlx_audio_separator/separator/models/demucs/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
+      "mlx_audio_separator/separator/models/mdx/__init__.py": "86849f8e4283b7182f02665a04d76955519a18f6b27fd8a58e5bd3a3c80f35eb",
+      "mlx_audio_separator/separator/models/mdx/convtdfnet.py": "672c64b1df1fc4fe66ac88cfb808e8972a03125009579641cfd200d7c3e5a937",
+      "mlx_audio_separator/separator/models/mdx/loader.py": "b83cb54906c9d79563555ca1cda97ef219ed498b4e2a7985d47daca052ffe3dd",
+      "mlx_audio_separator/separator/models/mdx/stft.py": "f9bae135a03e69741b7bede6e0ef330260c0d157d5f2d56bd5cad7c8a9a6171c",
+      "mlx_audio_separator/separator/models/mdxc/__init__.py": "d11e15564dd603d03d80426e63263c2b773c63bcacfcd7207818c4316830aba2",
+      "mlx_audio_separator/separator/models/mdxc/loader.py": "6d8dbc61bff3a28b107d3b2d715ddaf0a3cf85fad901bd318c54a6d1dd1aefec",
+      "mlx_audio_separator/separator/models/mdxc/tfc_tdf_v3_mlx.py": "a7d6534f4e718e982c037b50208a155cc9fcd31aecc738d1493df6d90091e79d",
+      "mlx_audio_separator/separator/models/roformer/__init__.py": "e7c8c7e8e35933e6563109cc519882bf95de7dc5a8e2e1cc9df1e783f7be81a2",
+      "mlx_audio_separator/separator/models/roformer/_core/__init__.py": "52fcea39466efad1116323236a51e61291aae5635887acdadba5fb2b11d39dce",
+      "mlx_audio_separator/separator/models/roformer/_core/config.py": "431e5f6c333c39c32d95a03521cc8f596d435f0e037ec0e0145565e3e5ef6b1b",
+      "mlx_audio_separator/separator/models/roformer/_core/convert.py": "66407e7e581633905642612d4c8f37db221b0f5a716ea016846c85af8baff485",
+      "mlx_audio_separator/separator/models/roformer/_core/kernels.py": "462bee4bb4f1bf93648549ba5345470a090c7c2e6e7be7fcaecfdd19c94ce11c",
+      "mlx_audio_separator/separator/models/roformer/_core/model.py": "fa8673f023fff1ba7f8a3ff0c64029cfc34a197d3df2eeb71cfaaad163129adc",
+      "mlx_audio_separator/separator/models/roformer/_core/ops.py": "5bbdc9c7918d269cb3710eceaa597356ecf6162dd5351dfe959a783520f519ce",
+      "mlx_audio_separator/separator/models/roformer/_core/separate.py": "0b1e73a66ce5e3b44e6b6eade1547c0d965a39425285c933e6d81d1ff6592542",
+      "mlx_audio_separator/separator/models/roformer/_core/stft.py": "528a2e9f6590d209aa3acaba9ccde028adc048b15aaf3e0ac673fe1a8de0351e",
+      "mlx_audio_separator/separator/models/roformer/_core/torch_checkpoint.py": "daa597f98ac64243516ba1b6297a31afbf6891db46ad809d54cb2862f04f5f32",
+      "mlx_audio_separator/separator/models/roformer/attention_kernel.py": "d9fff54cf1512eb8e2d55e9764643a311d04270e160b54778d607bc9c77f4bb2",
+      "mlx_audio_separator/separator/models/roformer/bs_loader.py": "22dcd6d31b21a28e5d035f6072ec8020617f8b032eaf49f75f8808199647d46d",
+      "mlx_audio_separator/separator/models/roformer/bs_roformer.py": "5cbfa4f57c0e76d96c4a2e27557bc8aadd67d8984d0dfff8350b5274ddf2409d",
+      "mlx_audio_separator/separator/models/roformer/components.py": "95e6618ee61fdcf2f09ec4ba4a2e3ac7deccbc3934f064f5fdb45018ea3b08df",
+      "mlx_audio_separator/separator/models/roformer/gelu.py": "35289288313436e239751cb6586d68c65dc032d115d2095147725508796d599a",
+      "mlx_audio_separator/separator/models/roformer/linear_kernel.py": "0164e0a0bc411816afbd07b593281b03d5aee75a286a3cd8878e456d59feb9a7",
+      "mlx_audio_separator/separator/models/roformer/loader.py": "e2323c97a6a232f17f8f4d8cca9a19712ef9fa31d14bffdd881e36099715c2d3",
+      "mlx_audio_separator/separator/models/roformer/mel_band_roformer.py": "4d052e729bd05304f8d19658ebb836fdf6ed9cd0952ce8700d0bfa1dd253da95",
+      "mlx_audio_separator/separator/models/roformer/metal_kernels.py": "7963bed370b16fb3b4084e4a8e6c073be5a6e75484f4ed69ce6b6bd15f717f4c",
+      "mlx_audio_separator/separator/models/roformer/overlap_add_kernels.py": "2c82fdab2530de985ff808330c6fb14dcfdeec71937691ea668e9d1ef091359a",
+      "mlx_audio_separator/separator/models/roformer/stft.py": "3101c957396af8280a3b916860556a945fd51afba4e6ad7ed18cff0a5587b457",
+      "mlx_audio_separator/separator/models/vr/__init__.py": "f07c510ea41b3bf8d192f3533294b3163c77a6744e9e19ffd89c3a045273c5cc",
+      "mlx_audio_separator/separator/models/vr/layers.py": "423f28f19805aaac834772e57377dd1f73a8d1786aa071d13124d442d6384145",
+      "mlx_audio_separator/separator/models/vr/loader.py": "1492693e6fed7ded4dd83df4d20d7ae4edc2789a96fb45c9f7bdaac97cbe0f49",
+      "mlx_audio_separator/separator/models/vr/nets.py": "164694d22fb08acbfafd3c6d5f6e87ccca6858d4226ebff8e091c0eef865e3a5",
+      "mlx_audio_separator/separator/models/vr/nets_new.py": "829d34393989afcf3583c9a44bc0a134d3da7e368183efdef15e8dd974703ec2",
+      "mlx_audio_separator/separator/models/vr/spec_utils.py": "4568a617d3ffe559e6931ae33666c1bc4c838c711749765f2d249d7167b79f7c",
+      "mlx_audio_separator/utils/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
+      "mlx_audio_separator/utils/benchmark.py": "b644d4405dbf001f6b8d14ddf7365d685ca353f0c05aac50c6281ccfd2fd5b7f",
+      "mlx_audio_separator/utils/cli.py": "e78b2aa10b42b9c14961f6c19530ad82ae38ffda17a18071efbe0f740e62b021",
+      "mlx_audio_separator/utils/equivalence.py": "083047b70c0d046886fc08fe9a42aa9001a1d3b031d09c2aa87de3c2c24bf2d7",
+      "mlx_audio_separator/utils/performance.py": "337d037ba418abe21d6bd6206d1f6c109a603c1155c9c4e01d8b2b52b4671a6a",
+      "scripts/catalog/catalog.py": "1af4c7ae15480b4c029291af7e34c89946dfcf82c54ee3f38a0ee3769f5ca06f",
+      "scripts/catalog/prepare.py": "8c7a740c0626848eaf2662cc48619a18458b6dc362ae0656567079c963f88486"
+    },
+    "versions": {
+      "mlx": "0.32.3",
+      "mlx-audio-io": "1.3.25",
+      "mlx-audio-separator": "0.1.21",
+      "mlx-spectro": "0.10.0",
+      "numpy": "2.5.3",
+      "onnx": "1.23.0",
+      "torch": "2.14.0"
+    },
+    "working_tree_modified": true
+  },
+  "distribution_layout": "single_hub_repository",
+  "format_version": 1,
+  "models": [
+    {
+      "architecture": "MDXC",
+      "artifact_origin": "converted",
+      "config_filename": "config.json",
+      "conversion_audit": {
+        "dtypes": [
+          "mlx.core.float32"
+        ],
+        "missing_parameters": [],
+        "shape_mismatches": [],
+        "tensor_count": 1939
+      },
+      "distribution": "bundled",
+      "dtype": "float32",
+      "family": "bs",
+      "files": [
+        {
+          "filename": "model.safetensors",
+          "sha256": "f95907cd5bfdbc783c1cb32f6aafc49b8b6a598477a9d4a48b4f9191c5167707",
+          "size_bytes": 698841589
+        },
+        {
+          "filename": "config.json",
+          "sha256": "386a86c2061536d2dca349943f1027f3ac5472ccb22b0e18bd528519b90f8931",
+          "size_bytes": 4282
+        }
+      ],
+      "id": "bs-roformer-sw",
+      "inference_defaults": {
+        "chunk_samples": 588800,
+        "overlap_windows": 2
+      },
+      "license": {
+        "note": "Source model card explicitly lists license: unknown.",
+        "status": "unresolved"
+      },
+      "name": "BS-RoFormer-SW",
+      "sample_rate": 44100,
+      "source": {
+        "config": {
+          "filename": "BS-Roformer-SW.yaml",
+          "sha256": "b558996f1e25eb48798bd6502505a5de94c4f966d6edfb1a0420f06cc40b501a",
+          "size_bytes": 4653
+        },
+        "download_url": "https://github.com/nomadkaraoke/python-audio-separator/releases/download/model-configs/BS-Roformer-SW.ckpt",
+        "upstream": "https://huggingface.co/enerjazzer/BS-ROFO-SW-Fixed/tree/a443a2985534b3bc815ef54a5d446c6a0390f974",
+        "weights": {
+          "filename": "BS-Roformer-SW.ckpt",
+          "sha256": "24e7d35ee9c64415673d3fd33e06a67cac2c103c5df6267ba1576459c775916e",
+          "size_bytes": 699412152
+        }
+      },
+      "stems": [
+        "bass",
+        "drums",
+        "other",
+        "vocals",
+        "guitar",
+        "piano"
+      ],
+      "validation": "validation/bs-roformer-sw.json",
+      "weights_filename": "model.safetensors"
+    },
+    {
+      "architecture": "MDXC",
+      "artifact_origin": "converted",
+      "config_filename": "config.json",
+      "conversion_audit": {
+        "dtypes": [
+          "mlx.core.float32"
+        ],
+        "missing_parameters": [],
+        "shape_mismatches": [],
+        "tensor_count": 672
+      },
+      "distribution": "bundled",
+      "dtype": "float32",
+      "family": "mel",
+      "files": [
+        {
+          "filename": "model.safetensors",
+          "sha256": "74397de5395ed372059b971d6398106f60d92373ada69b62e986e52c9697a5ba",
+          "size_bytes": 912889223
+        },
+        {
+          "filename": "config.json",
+          "sha256": "cd8548fd6a1345cfd0c51970919e15e7262cee954b6ff45bb016ed1364be38c5",
+          "size_bytes": 1150
+        }
+      ],
+      "id": "mel-roformer-instv7n",
+      "inference_defaults": {
+        "chunk_samples": 485100,
+        "overlap_windows": 2
+      },
+      "license": {
+        "note": "Weight redistribution terms must be confirmed; implementation license does not establish weight licensing.",
+        "status": "unresolved"
+      },
+      "name": "Mel-RoFormer InstV7n (Gabox)",
+      "sample_rate": 44100,
+      "source": {
+        "config": {
+          "filename": "config_mel_band_roformer_instrumental_gabox.yaml",
+          "sha256": "723af6755b5624be0a58351a13c930c472b51ef677cf2c7943394fefed7c3d4d",
+          "size_bytes": 920
+        },
+        "download_url": "https://github.com/nomadkaraoke/python-audio-separator/releases/download/model-configs/mel_band_roformer_instrumental_instv7n_gabox.ckpt",
+        "weights": {
+          "filename": "mel_band_roformer_instrumental_instv7n_gabox.ckpt",
+          "sha256": "b0ca36af5d1314be46b56c8a53b6be02f98511fa5d7e3e196fd895755e65be3c",
+          "size_bytes": 913026650
+        }
+      },
+      "stems": [
+        "Instrumental",
+        "Vocals"
+      ],
+      "validation": "validation/mel-roformer-instv7n.json",
+      "weights_filename": "model.safetensors"
+    },
+    {
+      "architecture": "MDX",
+      "artifact_origin": "converted",
+      "config_filename": "config.json",
+      "conversion_audit": {
+        "dtypes": [
+          "mlx.core.float32"
+        ],
+        "missing_parameters": [],
+        "shape_mismatches": [],
+        "tensor_count": 376
+      },
+      "distribution": "bundled",
+      "dtype": "float32",
+      "family": "mdx",
+      "files": [
+        {
+          "filename": "model.safetensors",
+          "sha256": "72c307239fbc415c9f3c1ef83fee38d19ac98b4b893f1a5afc4345296f48eea8",
+          "size_bytes": 66871394
+        },
+        {
+          "filename": "config.json",
+          "sha256": "d8df573c2b0248e3ce77eb8f0fc3c528c9db5368f38335e249e25e05e84415ba",
+          "size_bytes": 204
+        }
+      ],
+      "id": "uvr-mdx-net-inst-hq3",
+      "inference_defaults": {
+        "overlap_fraction": 0.25,
+        "segment_size": 256
+      },
+      "license": {
+        "note": "Weight redistribution terms must be confirmed; implementation license does not establish weight licensing.",
+        "status": "unresolved"
+      },
+      "name": "UVR-MDX-NET Inst HQ3",
+      "sample_rate": 44100,
+      "source": {
+        "config": {
+          "filename": "mdx_model_data.json",
+          "sha256": "1aca8f9bcc57233bc714029663a9ec2345d9c7721f91e5e08f46392a879c6a9a",
+          "size_bytes": 15358
+        },
+        "download_url": "https://github.com/TRvlvr/model_repo/releases/download/all_public_uvr_models/UVR-MDX-NET-Inst_HQ_3.onnx",
+        "weights": {
+          "filename": "UVR-MDX-NET-Inst_HQ_3.onnx",
+          "sha256": "317554b07fe1ea5279a77f2b1520a41ea4b93432560c4ffd08792c30fddf9adc",
+          "size_bytes": 66759214
+        }
+      },
+      "stems": [
+        "Instrumental",
+        "Vocals"
+      ],
+      "validation": "validation/uvr-mdx-net-inst-hq3.json",
+      "weights_filename": "model.safetensors"
+    },
+    {
+      "architecture": "Demucs",
+      "artifact_origin": "reused",
+      "config_filename": "htdemucs_config.json",
+      "distribution": "bundled",
+      "dtype": "float32",
+      "family": "demucs",
+      "files": [
+        {
+          "filename": "htdemucs.safetensors",
+          "sha256": "339d267a7a6983a11eedbdc00413c602a65e9b9103f695fb5c2b2a481cd9d297",
+          "size_bytes": 168005865
+        },
+        {
+          "filename": "htdemucs_config.json",
+          "sha256": "23657b19db14771aecf366ceedbf846c1386836c10043c97f4569acf05526b78",
+          "size_bytes": 4215
+        },
+        {
+          "filename": "LICENSE",
+          "sha256": "15086279d32c0f00c577c0f52ff428daf98b8a1fec0264da1c717c88ad464f51",
+          "size_bytes": 1117
+        }
+      ],
+      "id": "htdemucs",
+      "inference_defaults": {
+        "overlap_fraction": 0.25,
+        "shifts": 2
+      },
+      "license": {
+        "notice_source": "mlx_audio_separator/demucs_mlx/LICENSE",
+        "source": "https://huggingface.co/ssmall256/demucs-mlx/blob/35ecb1a906ec420b1a28069db575ad070d35860e/README.md",
+        "spdx": "MIT",
+        "status": "upstream_declared"
+      },
+      "name": "htdemucs",
+      "sample_rate": 44100,
+      "source": {
+        "note": "Reuses byte-identical published demucs-mlx artifacts; config records official checkpoint signatures and conversion metadata."
+      },
+      "stems": [
+        "drums",
+        "bass",
+        "other",
+        "vocals"
+      ],
+      "upstream_hub": {
+        "repo_id": "ssmall256/demucs-mlx",
+        "revision": "35ecb1a906ec420b1a28069db575ad070d35860e"
+      },
+      "validation": "validation/htdemucs.json",
+      "weights_filename": "htdemucs.safetensors"
+    },
+    {
+      "architecture": "Demucs",
+      "artifact_origin": "reused",
+      "config_filename": "htdemucs_ft_config.json",
+      "distribution": "bundled",
+      "dtype": "float32",
+      "family": "demucs",
+      "files": [
+        {
+          "filename": "htdemucs_ft.safetensors",
+          "sha256": "53f03b1ad4b4d211025a35da65460ba61a17547adf9c0544cad0ebcc8d7bbabb",
+          "size_bytes": 672024519
+        },
+        {
+          "filename": "htdemucs_ft_config.json",
+          "sha256": "6708535a698d4d4dd285a6f0a80d9d3c228fdd6514e53bdf1ddbc2ee9b0fa872",
+          "size_bytes": 10253
+        },
+        {
+          "filename": "LICENSE",
+          "sha256": "15086279d32c0f00c577c0f52ff428daf98b8a1fec0264da1c717c88ad464f51",
+          "size_bytes": 1117
+        }
+      ],
+      "id": "htdemucs_ft",
+      "inference_defaults": {
+        "overlap_fraction": 0.25,
+        "shifts": 2
+      },
+      "license": {
+        "notice_source": "mlx_audio_separator/demucs_mlx/LICENSE",
+        "source": "https://huggingface.co/ssmall256/demucs-mlx/blob/35ecb1a906ec420b1a28069db575ad070d35860e/README.md",
+        "spdx": "MIT",
+        "status": "upstream_declared"
+      },
+      "name": "htdemucs_ft",
+      "sample_rate": 44100,
+      "source": {
+        "note": "Reuses byte-identical published demucs-mlx artifacts; config records official checkpoint signatures and conversion metadata."
+      },
+      "stems": [
+        "drums",
+        "bass",
+        "other",
+        "vocals"
+      ],
+      "upstream_hub": {
+        "repo_id": "ssmall256/demucs-mlx",
+        "revision": "35ecb1a906ec420b1a28069db575ad070d35860e"
+      },
+      "validation": "validation/htdemucs_ft.json",
+      "weights_filename": "htdemucs_ft.safetensors"
+    },
+    {
+      "architecture": "Demucs",
+      "artifact_origin": "reused",
+      "config_filename": "htdemucs_6s_config.json",
+      "distribution": "bundled",
+      "dtype": "float32",
+      "family": "demucs",
+      "files": [
+        {
+          "filename": "htdemucs_6s.safetensors",
+          "sha256": "d298f7f746bf53c21baad44fb08e88807ef47feb551dd22f1601a546c85b8e02",
+          "size_bytes": 109726583
+        },
+        {
+          "filename": "htdemucs_6s_config.json",
+          "sha256": "d5e18c0209be583027d6eb5ec3013a02cd7ef55bc3096e644d8d3577d92e9bd0",
+          "size_bytes": 4302
+        },
+        {
+          "filename": "LICENSE",
+          "sha256": "15086279d32c0f00c577c0f52ff428daf98b8a1fec0264da1c717c88ad464f51",
+          "size_bytes": 1117
+        }
+      ],
+      "id": "htdemucs_6s",
+      "inference_defaults": {
+        "overlap_fraction": 0.25,
+        "shifts": 2
+      },
+      "license": {
+        "notice_source": "mlx_audio_separator/demucs_mlx/LICENSE",
+        "source": "https://huggingface.co/ssmall256/demucs-mlx/blob/35ecb1a906ec420b1a28069db575ad070d35860e/README.md",
+        "spdx": "MIT",
+        "status": "upstream_declared"
+      },
+      "name": "htdemucs_6s",
+      "sample_rate": 44100,
+      "source": {
+        "note": "Reuses byte-identical published demucs-mlx artifacts; config records official checkpoint signatures and conversion metadata."
+      },
+      "stems": [
+        "drums",
+        "bass",
+        "other",
+        "vocals",
+        "guitar",
+        "piano"
+      ],
+      "upstream_hub": {
+        "repo_id": "ssmall256/demucs-mlx",
+        "revision": "35ecb1a906ec420b1a28069db575ad070d35860e"
+      },
+      "validation": "validation/htdemucs_6s.json",
+      "weights_filename": "htdemucs_6s.safetensors"
+    },
+    {
+      "architecture": "MDXC",
+      "artifact_origin": "reused",
+      "config_filename": "config.json",
+      "distribution": "bundled",
+      "dtype": "float16",
+      "family": "zfturbo",
+      "files": [
+        {
+          "filename": "model.safetensors",
+          "sha256": "ef4aa052845a868cfaff93611477bd8f54d8081bc32f2742a9b3c738f0821191",
+          "size_bytes": 67402202
+        },
+        {
+          "filename": "config.json",
+          "sha256": "d110ed4713b403d27e2ae6c7796b1a864576232de4a2070e9762108ca197caa6",
+          "size_bytes": 1016
+        },
+        {
+          "filename": "LICENSE",
+          "sha256": "2b55ae02dcfe9ac8454c41c5e235175512db7429ee7275bab4c4e566b3f80597",
+          "size_bytes": 1779
+        }
+      ],
+      "id": "mel-roformer-zfturbo-vocals-v1",
+      "inference_defaults": {
+        "chunk_samples": 352800,
+        "overlap_windows": 2
+      },
+      "license": {
+        "note": "Published card notes the source MIT license was added after the original checkpoint release; explicit retroactive author confirmation was not collected.",
+        "notice_source": "https://huggingface.co/mlx-community/mel-roformer-zfturbo-vocals-v1-mlx/resolve/01bef2daebedf39937f552d6da68340fc979c9a0/LICENSE",
+        "source": "https://huggingface.co/mlx-community/mel-roformer-zfturbo-vocals-v1-mlx/blob/01bef2daebedf39937f552d6da68340fc979c9a0/README.md",
+        "spdx": "MIT",
+        "status": "upstream_declared_with_caveat"
+      },
+      "name": "ZFTurbo Mel-RoFormer Vocals v1",
+      "sample_rate": 44100,
+      "source": {
+        "_class": "MelRoFormerConfig",
+        "_dtype": "float16",
+        "_mlx_version": "0.31.0",
+        "_source_input": "model_vocals_mel_band_roformer_sdr_8.42.ckpt",
+        "_source_license_at_conversion": "MIT (current MSS-Training repo LICENSE; LICENSE file added 2024-11-04 in commit 6149a22, post-dating the v1.0.0 release on 2023-11-06)",
+        "_source_release": "https://github.com/ZFTurbo/Music-Source-Separation-Training/releases/tag/v1.0.0",
+        "_source_repository": "https://github.com/ZFTurbo/Music-Source-Separation-Training",
+        "_source_sha256": "d9ce706b49cebf0af018590d8deb47ad5434987bf8f7bd3a87a4e5e8c30acb26",
+        "_tool_version": "8380ab8",
+        "checkpoint_family": "zfturbo_vocals_v1",
+        "chunk_size": 352800,
+        "depth": 8,
+        "dim": 192,
+        "dim_head": 64,
+        "ff_mult": 4,
+        "heads": 8,
+        "hop_length": 512,
+        "mask_estimator_depth": 1,
+        "mlp_expansion_factor": 4,
+        "model_type": "mel_band_roformer",
+        "n_fft": 2048,
+        "num_bands": 60,
+        "num_overlap": 2,
+        "num_stems": 1,
+        "sample_rate": 44100,
+        "win_length": 2048
+      },
+      "stems": [
+        "Vocals",
+        "Instrumental"
+      ],
+      "upstream_hub": {
+        "repo_id": "mlx-community/mel-roformer-zfturbo-vocals-v1-mlx",
+        "revision": "01bef2daebedf39937f552d6da68340fc979c9a0"
+      },
+      "validation": "validation/mel-roformer-zfturbo-vocals-v1.json",
+      "weights_filename": "model.safetensors"
+    }
+  ],
+  "publication": {
+    "date": "2026-10-11",
+    "minimum_runtime_version": "0.1.21",
+    "repository": "ssmall256/mlx-audio-separator",
+    "runtime_release_status": "upcoming_at_publication",
+    "third_party_notices": "THIRD_PARTY_NOTICES.md"
+  },
+  "review_tools_sha256": {
+    "catalog.py": "41469a84c9cd1ea89be7cc98eae2e72837e5b8d897b9d3fc368d1538015a5c14",
+    "export_registry.py": "b688c030d8c8d14113bc9f0b9fbba97723a42ad2d847b6e2ce9725bda0e37653",
+    "finalize.py": "65208c68de24e2520bbdbaef6c8930f8f590b4ac3e44206d85c8e79c59f1b782",
+    "prepare.py": "81ca71f53e24aa53eefc4b03beb563c84a7df14e4b239adb71eec5d802a88df5",
+    "preview.py": "1fdd89ad568d206c103e176e363d3abcd2a508106f94425cc2b59341df07c42c",
+    "validate.py": "aca2ea18ca56dd71602d5313c825591c87f133e8cccf97266ef672327024e35f",
+    "validate_downloads.py": "7e38c0d1678235ad340d62f302c25588ae29fe98b526806089da68235da59f4d"
+  },
+  "target_repo_id": "ssmall256/mlx-audio-separator",
+  "validation_scope": {
+    "artifact_inference_job": "mq-07b0fe",
+    "artifact_relocation_changed_weights": false,
+    "baseline_job": "mq-a8c83e",
+    "conversion_job": "mq-565cea",
+    "current_bundle_all_models_hash_verified": 7,
+    "fixture": "14-second excerpt of Q-BHR.m4a, starting at 45 seconds, decoded once to stereo 44.1 kHz PCM16",
+    "fixture_not_distributed": true,
+    "fixture_sha256": "f5d0902f372a61ae2784239d0e097c9246c9d9b90be26a4a4805db8c96efad6d",
+    "frames": 617400,
+    "manifest_job": "mq-15ec6e",
+    "new_conversions_materialized_into_empty_cache": 3,
+    "new_ground_truth_quality_evaluation": false,
+    "new_torch_reference_comparison": false,
+    "reused_artifacts_downloaded_from_upstream_into_empty_cache": 4
+  },
+  "validation_status": "passed_artifact_integrity_offline_inference_and_conversion_roundtrip"
+}

catalog.py

@@ -0,0 +1,188 @@
+"""Verify, materialize, and run a curated catalog. This tool never uploads files."""
+
+from __future__ import annotations
+
+import argparse
+import hashlib
+import json
+import os
+import shutil
+import tempfile
+import urllib.request
+from pathlib import Path
+
+
+def digest(path):
+    h = hashlib.sha256()
+    with Path(path).open("rb") as stream:
+        for chunk in iter(lambda: stream.read(4 << 20), b""):
+            h.update(chunk)
+    return h.hexdigest()
+
+
+def file_record(path, filename=None):
+    path = Path(path)
+    return {"filename": filename or path.name, "size_bytes": path.stat().st_size, "sha256": digest(path)}
+
+
+def safe_path(root, relative):
+    root = Path(root).resolve()
+    path = (root / relative).resolve()
+    if not path.is_relative_to(root) or path == root:
+        raise ValueError(f"Path escapes catalog directory: {relative}")
+    return path
+
+
+def verify_file(path, record):
+    path = Path(path)
+    if not path.is_file() or path.stat().st_size != record["size_bytes"] or digest(path) != record["sha256"]:
+        raise ValueError(f"Size or SHA-256 mismatch: {path}")
+
+
+def read_manifest(bundle):
+    manifest = json.loads((Path(bundle) / "catalog.json").read_text())
+    if manifest["format_version"] != 1:
+        raise ValueError("Unsupported catalog version")
+    ids = [entry["id"] for entry in manifest["models"]]
+    if len(ids) != len(set(ids)):
+        raise ValueError("Duplicate catalog model IDs")
+    return manifest
+
+
+def materialize(bundle, cache, selected=None):
+    """Install verified files atomically; never fetch source .ckpt/.onnx files."""
+    bundle, cache = Path(bundle), Path(cache)
+    manifest = read_manifest(bundle)
+    unknown = set(selected or []) - {entry["id"] for entry in manifest["models"]}
+    if unknown:
+        raise ValueError(f"Unknown catalog models: {sorted(unknown)}")
+    for entry in manifest["models"]:
+        if selected and entry["id"] not in selected:
+            continue
+        directory = safe_path(cache, entry["id"])
+        directory.mkdir(parents=True, exist_ok=True)
+        for record in entry["files"]:
+            target = safe_path(directory, record["filename"])
+            try:
+                verify_file(target, record)
+                continue
+            except ValueError:
+                pass
+            temporary = None
+            try:
+                target.parent.mkdir(parents=True, exist_ok=True)
+                with tempfile.NamedTemporaryFile(dir=target.parent, prefix=".download-", delete=False) as out:
+                    temporary = Path(out.name)
+                    if entry["distribution"] == "bundled":
+                        source = safe_path(bundle, f"models/{entry['id']}/{record['filename']}")
+                        with source.open("rb") as stream:
+                            shutil.copyfileobj(stream, out, 4 << 20)
+                    else:
+                        revision = entry["revision"]
+                        if len(revision) != 40 or any(c not in "0123456789abcdef" for c in revision):
+                            raise ValueError("Remote entries require an immutable commit SHA")
+                        url = f"https://huggingface.co/{entry['repo_id']}/resolve/{revision}/{record['filename']}"
+                        with urllib.request.urlopen(url, timeout=120) as stream:
+                            shutil.copyfileobj(stream, out, 4 << 20)
+                verify_file(temporary, record)
+                temporary.replace(target)
+            finally:
+                if temporary:
+                    temporary.unlink(missing_ok=True)
+        print(f"Verified {entry['id']}", flush=True)
+    return manifest
+
+
+def bundle_published_artifacts(bundle):
+    """Copy all pinned artifacts into one self-contained publication directory."""
+    bundle = Path(bundle)
+    manifest = materialize(bundle, bundle / "models")
+    for entry in manifest["models"]:
+        if entry["distribution"] == "external":
+            entry["artifact_origin"] = "reused"
+            entry["upstream_hub"] = {"repo_id": entry.pop("repo_id"), "revision": entry.pop("revision")}
+        else:
+            entry.setdefault("artifact_origin", "converted")
+        entry["distribution"] = "bundled"
+    manifest["target_repo_id"] = "ssmall256/mlx-audio-separator"
+    manifest["distribution_layout"] = "single_hub_repository"
+    temporary = bundle / "catalog.json.tmp"
+    temporary.write_text(json.dumps(manifest, indent=2, sort_keys=True) + "\n")
+    temporary.replace(bundle / "catalog.json")
+    return manifest
+
+
+def local_separator(entry, directory, output):
+    """Use the release's normal separation path with catalog metadata and local weights."""
+    from mlx_audio_separator import Separator
+
+    directory = Path(directory)
+    if entry["family"] == "demucs":
+        os.environ["DEMUCS_MLX_CACHE_DIR"] = str(directory)
+        os.environ["DEMUCS_MLX_NO_DOWNLOAD"] = "1"
+    os.environ["HF_HUB_OFFLINE"] = "1"
+
+    class CatalogSeparator(Separator):
+        def download_model_files(self, model_filename):
+            path = directory if entry["family"] == "zfturbo" else directory / entry["weights_filename"]
+            return entry["id"], entry["architecture"], entry["name"], str(path), None
+
+        def _load_downloaded_model_data(self, model_path, yaml_config_filename):
+            if entry["family"] == "zfturbo":
+                from mlx_audio_separator.hf_mel_roformer import load_model_data
+
+                return load_model_data(directory)
+            if entry["family"] == "demucs":
+                return {"name": entry["id"]}
+            return json.loads((directory / entry["config_filename"]).read_text())
+
+    separator = CatalogSeparator(
+        model_file_dir=str(directory),
+        output_dir=str(output),
+        log_level=30,
+        demucs_params={"shifts": 2, "seed": 0},
+    )
+    separator.load_model(entry["id"])
+    return separator
+
+
+def run(bundle, cache, model, audio, output):
+    manifest = read_manifest(bundle)
+    entry = next(entry for entry in manifest["models"] if entry["id"] == model)
+    directory = safe_path(cache, model)
+    for record in entry["files"]:
+        verify_file(safe_path(directory, record["filename"]), record)
+    separator = local_separator(entry, directory, output)
+    results = separator.separate(str(audio))
+    if separator.failed_files or len(results) != len(entry["stems"]):
+        raise RuntimeError(f"Incomplete separation: {separator.failed_files}, {results}")
+    return results
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("command", choices=["verify", "fetch", "separate"])
+    parser.add_argument("--bundle", type=Path, default=Path(__file__).resolve().parent)
+    parser.add_argument("--cache", type=Path, default=Path.home() / ".cache/mlx-audio-separator/curated-v1")
+    parser.add_argument("--model")
+    parser.add_argument("--audio", type=Path)
+    parser.add_argument("--output", type=Path, default=Path("stems"))
+    args = parser.parse_args()
+    if args.command == "fetch":
+        materialize(args.bundle, args.cache, [args.model] if args.model else None)
+    elif args.command == "verify":
+        manifest = read_manifest(args.bundle)
+        for entry in manifest["models"]:
+            directory = args.bundle / "models" / entry["id"] if entry["distribution"] == "bundled" else args.cache / entry["id"]
+            for record in entry["files"]:
+                verify_file(safe_path(directory, record["filename"]), record)
+            print(f"Verified {entry['id']}")
+    else:
+        if not args.model or not args.audio:
+            parser.error("separate requires --model and --audio")
+        for path in run(args.bundle, args.cache, args.model, args.audio, args.output):
+            print(path)
+
+
+if __name__ == "__main__":
+    main()

models/bs-roformer-sw/config.json

@@ -0,0 +1,211 @@
+{
+  "audio": {
+    "chunk_size": 588800,
+    "dim_f": 1024,
+    "dim_t": 801,
+    "hop_length": 441,
+    "min_mean_abs": 0.0,
+    "n_fft": 2048,
+    "num_channels": 2,
+    "sample_rate": 44100
+  },
+  "augmentations": {
+    "all": {
+      "channel_shuffle": 0.5,
+      "random_inverse": 0.1,
+      "random_polarity": 0.5
+    },
+    "bass": {
+      "pitch_shift": 0.1,
+      "pitch_shift_max_semitones": 2,
+      "pitch_shift_min_semitones": -2,
+      "seven_band_parametric_eq": 0.1,
+      "seven_band_parametric_eq_max_gain_db": 6,
+      "seven_band_parametric_eq_min_gain_db": -3,
+      "tanh_distortion": 0.1,
+      "tanh_distortion_max": 0.5,
+      "tanh_distortion_min": 0.1
+    },
+    "drums": {
+      "pitch_shift": 0.1,
+      "pitch_shift_max_semitones": 5,
+      "pitch_shift_min_semitones": -5,
+      "seven_band_parametric_eq": 0.1,
+      "seven_band_parametric_eq_max_gain_db": 9,
+      "seven_band_parametric_eq_min_gain_db": -9,
+      "tanh_distortion": 0.1,
+      "tanh_distortion_max": 0.6,
+      "tanh_distortion_min": 0.1
+    },
+    "enable": true,
+    "loudness": true,
+    "loudness_max": 1.5,
+    "loudness_min": 0.5,
+    "mixup": true,
+    "mixup_loudness_max": 1.5,
+    "mixup_loudness_min": 0.5,
+    "mixup_probs": [
+      0.2,
+      0.02
+    ],
+    "other": {
+      "gaussian_noise": 0.1,
+      "gaussian_noise_max_amplitude": 0.015,
+      "gaussian_noise_min_amplitude": 0.001,
+      "pitch_shift": 0.1,
+      "pitch_shift_max_semitones": 4,
+      "pitch_shift_min_semitones": -4,
+      "time_stretch": 0.1,
+      "time_stretch_max_rate": 1.25,
+      "time_stretch_min_rate": 0.8
+    },
+    "vocals": {
+      "pitch_shift": 0.1,
+      "pitch_shift_max_semitones": 5,
+      "pitch_shift_min_semitones": -5,
+      "seven_band_parametric_eq": 0.1,
+      "seven_band_parametric_eq_max_gain_db": 9,
+      "seven_band_parametric_eq_min_gain_db": -9,
+      "tanh_distortion": 0.1,
+      "tanh_distortion_max": 0.7,
+      "tanh_distortion_min": 0.1
+    }
+  },
+  "inference": {
+    "batch_size": 1,
+    "dim_t": 801,
+    "normalize": false,
+    "num_overlap": 2
+  },
+  "is_roformer": true,
+  "model": {
+    "attn_dropout": 0.1,
+    "depth": 12,
+    "dim": 256,
+    "dim_freqs_in": 1025,
+    "dim_head": 64,
+    "ff_dropout": 0.1,
+    "flash_attn": true,
+    "freq_transformer_depth": 1,
+    "freqs_per_bands": [
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      2,
+      4,
+      4,
+      4,
+      4,
+      4,
+      4,
+      4,
+      4,
+      4,
+      4,
+      4,
+      4,
+      12,
+      12,
+      12,
+      12,
+      12,
+      12,
+      12,
+      12,
+      24,
+      24,
+      24,
+      24,
+      24,
+      24,
+      24,
+      24,
+      48,
+      48,
+      48,
+      48,
+      48,
+      48,
+      48,
+      48,
+      128,
+      129
+    ],
+    "heads": 8,
+    "linear_transformer_depth": 0,
+    "mask_estimator_depth": 2,
+    "mlp_expansion_factor": 4,
+    "multi_stft_hop_size": 147,
+    "multi_stft_normalized": false,
+    "multi_stft_resolution_loss_weight": 1.0,
+    "multi_stft_resolutions_window_sizes": [
+      4096,
+      2048,
+      1024,
+      512,
+      256
+    ],
+    "num_stems": 6,
+    "skip_connection": false,
+    "stereo": true,
+    "stft_hop_length": 512,
+    "stft_n_fft": 2048,
+    "stft_normalized": false,
+    "stft_win_length": 2048,
+    "time_transformer_depth": 1,
+    "use_torch_checkpoint": false
+  },
+  "training": {
+    "augmentation": false,
+    "augmentation_loudness": true,
+    "augmentation_loudness_max": 1.5,
+    "augmentation_loudness_min": 0.5,
+    "augmentation_loudness_type": 1,
+    "augmentation_mix": true,
+    "augmentation_type": "simple1",
+    "batch_size": 2,
+    "coarse_loss_clip": true,
+    "ema_momentum": 0.999,
+    "grad_clip": 0,
+    "gradient_accumulation_steps": 1,
+    "instruments": [
+      "bass",
+      "drums",
+      "other",
+      "vocals",
+      "guitar",
+      "piano"
+    ],
+    "lr": 1e-05,
+    "num_epochs": 1000,
+    "num_steps": 1000,
+    "optimizer": "adam",
+    "other_fix": false,
+    "patience": 3,
+    "q": 0.95,
+    "reduce_factor": 0.95,
+    "target_instrument": null,
+    "use_amp": true,
+    "use_mp3_compress": false
+  }
+}

models/bs-roformer-sw/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f95907cd5bfdbc783c1cb32f6aafc49b8b6a598477a9d4a48b4f9191c5167707
+size 698841589

models/htdemucs/LICENSE

@@ -0,0 +1,22 @@
+MIT License
+
+Copyright (c) Meta Platforms, Inc. and affiliates.
+Copyright (c) 2026 ssmall256
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

models/htdemucs/htdemucs.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:339d267a7a6983a11eedbdc00413c602a65e9b9103f695fb5c2b2a481cd9d297
+size 168005865

models/htdemucs/htdemucs_config.json

@@ -0,0 +1,188 @@
+{
+  "args": [],
+  "conversion_date": "2026-10-05T14:40:20.987896+00:00",
+  "format_version": 1,
+  "kwargs": {
+    "audio_channels": 2,
+    "bottom_channels": 512,
+    "cac": true,
+    "channels": 48,
+    "channels_time": null,
+    "context": 1,
+    "context_enc": 0,
+    "dconv_comp": 8,
+    "dconv_depth": 2,
+    "dconv_init": 0.001,
+    "dconv_mode": 3,
+    "depth": 4,
+    "emb_scale": 10,
+    "emb_smooth": true,
+    "end_iters": 0,
+    "freq_emb": 0.2,
+    "growth": 2,
+    "kernel_size": 8,
+    "multi_freqs": [],
+    "multi_freqs_depth": 3,
+    "nfft": 4096,
+    "norm_groups": 4,
+    "norm_starts": 4,
+    "rescale": 0.1,
+    "rewrite": true,
+    "samplerate": 44100,
+    "segment": {
+      "__type__": "fraction",
+      "denominator": 5,
+      "numerator": 39
+    },
+    "sources": [
+      "drums",
+      "bass",
+      "other",
+      "vocals"
+    ],
+    "stride": 4,
+    "t_auto_sparsity": false,
+    "t_cape_augment": true,
+    "t_cape_glob_loc_scale": [
+      5000.0,
+      1.0,
+      1.4
+    ],
+    "t_cape_mean_normalize": true,
+    "t_cross_first": false,
+    "t_dropout": 0.02,
+    "t_emb": "sin",
+    "t_gelu": true,
+    "t_global_window": 100,
+    "t_group_norm": false,
+    "t_heads": 8,
+    "t_hidden_scale": 4.0,
+    "t_layer_scale": true,
+    "t_layers": 5,
+    "t_lr": null,
+    "t_mask_random_seed": 42,
+    "t_mask_type": "diag",
+    "t_max_period": 10000.0,
+    "t_max_positions": 10000,
+    "t_norm_first": true,
+    "t_norm_in": true,
+    "t_norm_in_group": false,
+    "t_norm_out": true,
+    "t_sin_random_shift": 0,
+    "t_sparse_attn_window": 400,
+    "t_sparse_cross_attn": false,
+    "t_sparse_self_attn": false,
+    "t_sparsity": 0.95,
+    "t_weight_decay": 0.0,
+    "t_weight_pos_embed": 1.0,
+    "time_stride": 2,
+    "wiener_iters": 0,
+    "wiener_residual": false
+  },
+  "mlx_version": "0.32.3",
+  "model_class": "BagOfModelsMLX",
+  "model_name": "htdemucs",
+  "num_models": 1,
+  "per_model_args": [
+    []
+  ],
+  "per_model_classes": [
+    "HTDemucsMLX"
+  ],
+  "per_model_kwargs": [
+    {
+      "audio_channels": 2,
+      "bottom_channels": 512,
+      "cac": true,
+      "channels": 48,
+      "channels_time": null,
+      "context": 1,
+      "context_enc": 0,
+      "dconv_comp": 8,
+      "dconv_depth": 2,
+      "dconv_init": 0.001,
+      "dconv_mode": 3,
+      "depth": 4,
+      "emb_scale": 10,
+      "emb_smooth": true,
+      "end_iters": 0,
+      "freq_emb": 0.2,
+      "growth": 2,
+      "kernel_size": 8,
+      "multi_freqs": [],
+      "multi_freqs_depth": 3,
+      "nfft": 4096,
+      "norm_groups": 4,
+      "norm_starts": 4,
+      "rescale": 0.1,
+      "rewrite": true,
+      "samplerate": 44100,
+      "segment": {
+        "__type__": "fraction",
+        "denominator": 5,
+        "numerator": 39
+      },
+      "sources": [
+        "drums",
+        "bass",
+        "other",
+        "vocals"
+      ],
+      "stride": 4,
+      "t_auto_sparsity": false,
+      "t_cape_augment": true,
+      "t_cape_glob_loc_scale": [
+        5000.0,
+        1.0,
+        1.4
+      ],
+      "t_cape_mean_normalize": true,
+      "t_cross_first": false,
+      "t_dropout": 0.02,
+      "t_emb": "sin",
+      "t_gelu": true,
+      "t_global_window": 100,
+      "t_group_norm": false,
+      "t_heads": 8,
+      "t_hidden_scale": 4.0,
+      "t_layer_scale": true,
+      "t_layers": 5,
+      "t_lr": null,
+      "t_mask_random_seed": 42,
+      "t_mask_type": "diag",
+      "t_max_period": 10000.0,
+      "t_max_positions": 10000,
+      "t_norm_first": true,
+      "t_norm_in": true,
+      "t_norm_in_group": false,
+      "t_norm_out": true,
+      "t_sin_random_shift": 0,
+      "t_sparse_attn_window": 400,
+      "t_sparse_cross_attn": false,
+      "t_sparse_self_attn": false,
+      "t_sparsity": 0.95,
+      "t_weight_decay": 0.0,
+      "t_weight_pos_embed": 1.0,
+      "time_stride": 2,
+      "wiener_iters": 0,
+      "wiener_residual": false
+    }
+  ],
+  "safetensors_sha256": "339d267a7a6983a11eedbdc00413c602a65e9b9103f695fb5c2b2a481cd9d297",
+  "source_artifacts": [
+    {
+      "checksum": "8726e21a",
+      "signature": "955717e8"
+    }
+  ],
+  "sub_model_class": "HTDemucsMLX",
+  "verification_passed": true,
+  "weights": [
+    [
+      1.0,
+      1.0,
+      1.0,
+      1.0
+    ]
+  ]
+}

models/htdemucs_6s/LICENSE

@@ -0,0 +1,22 @@
+MIT License
+
+Copyright (c) Meta Platforms, Inc. and affiliates.
+Copyright (c) 2026 ssmall256
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

models/htdemucs_6s/htdemucs_6s.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d298f7f746bf53c21baad44fb08e88807ef47feb551dd22f1601a546c85b8e02
+size 109726583

models/htdemucs_6s/htdemucs_6s_config.json

@@ -0,0 +1,194 @@
+{
+  "args": [],
+  "conversion_date": "2026-10-05T14:40:32.682284+00:00",
+  "format_version": 1,
+  "kwargs": {
+    "audio_channels": 2,
+    "bottom_channels": 0,
+    "cac": true,
+    "channels": 48,
+    "channels_time": null,
+    "context": 1,
+    "context_enc": 0,
+    "dconv_comp": 8,
+    "dconv_depth": 2,
+    "dconv_init": 0.001,
+    "dconv_mode": 3,
+    "depth": 4,
+    "emb_scale": 10,
+    "emb_smooth": true,
+    "end_iters": 0,
+    "freq_emb": 0.2,
+    "growth": 2,
+    "kernel_size": 8,
+    "multi_freqs": [],
+    "multi_freqs_depth": 3,
+    "nfft": 4096,
+    "norm_groups": 4,
+    "norm_starts": 4,
+    "rescale": 0.1,
+    "rewrite": true,
+    "samplerate": 44100,
+    "segment": {
+      "__type__": "fraction",
+      "denominator": 5,
+      "numerator": 39
+    },
+    "sources": [
+      "drums",
+      "bass",
+      "other",
+      "vocals",
+      "guitar",
+      "piano"
+    ],
+    "stride": 4,
+    "t_auto_sparsity": false,
+    "t_cape_augment": true,
+    "t_cape_glob_loc_scale": [
+      5000.0,
+      1.0,
+      1.4
+    ],
+    "t_cape_mean_normalize": true,
+    "t_cross_first": false,
+    "t_dropout": 0.02,
+    "t_emb": "sin",
+    "t_gelu": true,
+    "t_global_window": 100,
+    "t_group_norm": false,
+    "t_heads": 8,
+    "t_hidden_scale": 4.0,
+    "t_layer_scale": true,
+    "t_layers": 5,
+    "t_lr": null,
+    "t_mask_random_seed": 42,
+    "t_mask_type": "diag",
+    "t_max_period": 10000.0,
+    "t_max_positions": 10000,
+    "t_norm_first": true,
+    "t_norm_in": true,
+    "t_norm_in_group": false,
+    "t_norm_out": true,
+    "t_sin_random_shift": 0,
+    "t_sparse_attn_window": 400,
+    "t_sparse_cross_attn": false,
+    "t_sparse_self_attn": false,
+    "t_sparsity": 0.95,
+    "t_weight_decay": 0.0,
+    "t_weight_pos_embed": 1.0,
+    "time_stride": 2,
+    "wiener_iters": 0,
+    "wiener_residual": false
+  },
+  "mlx_version": "0.32.3",
+  "model_class": "BagOfModelsMLX",
+  "model_name": "htdemucs_6s",
+  "num_models": 1,
+  "per_model_args": [
+    []
+  ],
+  "per_model_classes": [
+    "HTDemucsMLX"
+  ],
+  "per_model_kwargs": [
+    {
+      "audio_channels": 2,
+      "bottom_channels": 0,
+      "cac": true,
+      "channels": 48,
+      "channels_time": null,
+      "context": 1,
+      "context_enc": 0,
+      "dconv_comp": 8,
+      "dconv_depth": 2,
+      "dconv_init": 0.001,
+      "dconv_mode": 3,
+      "depth": 4,
+      "emb_scale": 10,
+      "emb_smooth": true,
+      "end_iters": 0,
+      "freq_emb": 0.2,
+      "growth": 2,
+      "kernel_size": 8,
+      "multi_freqs": [],
+      "multi_freqs_depth": 3,
+      "nfft": 4096,
+      "norm_groups": 4,
+      "norm_starts": 4,
+      "rescale": 0.1,
+      "rewrite": true,
+      "samplerate": 44100,
+      "segment": {
+        "__type__": "fraction",
+        "denominator": 5,
+        "numerator": 39
+      },
+      "sources": [
+        "drums",
+        "bass",
+        "other",
+        "vocals",
+        "guitar",
+        "piano"
+      ],
+      "stride": 4,
+      "t_auto_sparsity": false,
+      "t_cape_augment": true,
+      "t_cape_glob_loc_scale": [
+        5000.0,
+        1.0,
+        1.4
+      ],
+      "t_cape_mean_normalize": true,
+      "t_cross_first": false,
+      "t_dropout": 0.02,
+      "t_emb": "sin",
+      "t_gelu": true,
+      "t_global_window": 100,
+      "t_group_norm": false,
+      "t_heads": 8,
+      "t_hidden_scale": 4.0,
+      "t_layer_scale": true,
+      "t_layers": 5,
+      "t_lr": null,
+      "t_mask_random_seed": 42,
+      "t_mask_type": "diag",
+      "t_max_period": 10000.0,
+      "t_max_positions": 10000,
+      "t_norm_first": true,
+      "t_norm_in": true,
+      "t_norm_in_group": false,
+      "t_norm_out": true,
+      "t_sin_random_shift": 0,
+      "t_sparse_attn_window": 400,
+      "t_sparse_cross_attn": false,
+      "t_sparse_self_attn": false,
+      "t_sparsity": 0.95,
+      "t_weight_decay": 0.0,
+      "t_weight_pos_embed": 1.0,
+      "time_stride": 2,
+      "wiener_iters": 0,
+      "wiener_residual": false
+    }
+  ],
+  "safetensors_sha256": "d298f7f746bf53c21baad44fb08e88807ef47feb551dd22f1601a546c85b8e02",
+  "source_artifacts": [
+    {
+      "checksum": "34c22ccb",
+      "signature": "5c90dfd2"
+    }
+  ],
+  "sub_model_class": "HTDemucsMLX",
+  "verification_passed": true,
+  "weights": [
+    [
+      1.0,
+      1.0,
+      1.0,
+      1.0,
+      1.0,
+      1.0
+    ]
+  ]
+}

models/htdemucs_ft/LICENSE

@@ -0,0 +1,22 @@
+MIT License
+
+Copyright (c) Meta Platforms, Inc. and affiliates.
+Copyright (c) 2026 ssmall256
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

models/htdemucs_ft/htdemucs_ft.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53f03b1ad4b4d211025a35da65460ba61a17547adf9c0544cad0ebcc8d7bbabb
+size 672024519

models/htdemucs_ft/htdemucs_ft_config.json

@@ -0,0 +1,455 @@
+{
+  "args": [],
+  "conversion_date": "2026-10-05T14:40:24.444402+00:00",
+  "format_version": 1,
+  "kwargs": {
+    "audio_channels": 2,
+    "bottom_channels": 512,
+    "cac": true,
+    "channels": 48,
+    "channels_time": null,
+    "context": 1,
+    "context_enc": 0,
+    "dconv_comp": 8,
+    "dconv_depth": 2,
+    "dconv_init": 0.001,
+    "dconv_mode": 3,
+    "depth": 4,
+    "emb_scale": 10,
+    "emb_smooth": true,
+    "end_iters": 0,
+    "freq_emb": 0.2,
+    "growth": 2,
+    "kernel_size": 8,
+    "multi_freqs": [],
+    "multi_freqs_depth": 3,
+    "nfft": 4096,
+    "norm_groups": 4,
+    "norm_starts": 4,
+    "rescale": 0.1,
+    "rewrite": true,
+    "samplerate": 44100,
+    "segment": {
+      "__type__": "fraction",
+      "denominator": 5,
+      "numerator": 39
+    },
+    "sources": [
+      "drums",
+      "bass",
+      "other",
+      "vocals"
+    ],
+    "stride": 4,
+    "t_auto_sparsity": false,
+    "t_cape_augment": true,
+    "t_cape_glob_loc_scale": [
+      5000.0,
+      1.0,
+      1.4
+    ],
+    "t_cape_mean_normalize": true,
+    "t_cross_first": false,
+    "t_dropout": 0.02,
+    "t_emb": "sin",
+    "t_gelu": true,
+    "t_global_window": 100,
+    "t_group_norm": false,
+    "t_heads": 8,
+    "t_hidden_scale": 4.0,
+    "t_layer_scale": true,
+    "t_layers": 5,
+    "t_lr": null,
+    "t_mask_random_seed": 42,
+    "t_mask_type": "diag",
+    "t_max_period": 10000.0,
+    "t_max_positions": 10000,
+    "t_norm_first": true,
+    "t_norm_in": true,
+    "t_norm_in_group": false,
+    "t_norm_out": true,
+    "t_sin_random_shift": 0,
+    "t_sparse_attn_window": 400,
+    "t_sparse_cross_attn": false,
+    "t_sparse_self_attn": false,
+    "t_sparsity": 0.95,
+    "t_weight_decay": 0.05,
+    "t_weight_pos_embed": 1.0,
+    "time_stride": 2,
+    "wiener_iters": 0,
+    "wiener_residual": false
+  },
+  "mlx_version": "0.32.3",
+  "model_class": "BagOfModelsMLX",
+  "model_name": "htdemucs_ft",
+  "num_models": 4,
+  "per_model_args": [
+    [],
+    [],
+    [],
+    []
+  ],
+  "per_model_classes": [
+    "HTDemucsMLX",
+    "HTDemucsMLX",
+    "HTDemucsMLX",
+    "HTDemucsMLX"
+  ],
+  "per_model_kwargs": [
+    {
+      "audio_channels": 2,
+      "bottom_channels": 512,
+      "cac": true,
+      "channels": 48,
+      "channels_time": null,
+      "context": 1,
+      "context_enc": 0,
+      "dconv_comp": 8,
+      "dconv_depth": 2,
+      "dconv_init": 0.001,
+      "dconv_mode": 3,
+      "depth": 4,
+      "emb_scale": 10,
+      "emb_smooth": true,
+      "end_iters": 0,
+      "freq_emb": 0.2,
+      "growth": 2,
+      "kernel_size": 8,
+      "multi_freqs": [],
+      "multi_freqs_depth": 3,
+      "nfft": 4096,
+      "norm_groups": 4,
+      "norm_starts": 4,
+      "rescale": 0.1,
+      "rewrite": true,
+      "samplerate": 44100,
+      "segment": {
+        "__type__": "fraction",
+        "denominator": 5,
+        "numerator": 39
+      },
+      "sources": [
+        "drums",
+        "bass",
+        "other",
+        "vocals"
+      ],
+      "stride": 4,
+      "t_auto_sparsity": false,
+      "t_cape_augment": true,
+      "t_cape_glob_loc_scale": [
+        5000.0,
+        1.0,
+        1.4
+      ],
+      "t_cape_mean_normalize": true,
+      "t_cross_first": false,
+      "t_dropout": 0.02,
+      "t_emb": "sin",
+      "t_gelu": true,
+      "t_global_window": 100,
+      "t_group_norm": false,
+      "t_heads": 8,
+      "t_hidden_scale": 4.0,
+      "t_layer_scale": true,
+      "t_layers": 5,
+      "t_lr": null,
+      "t_mask_random_seed": 42,
+      "t_mask_type": "diag",
+      "t_max_period": 10000.0,
+      "t_max_positions": 10000,
+      "t_norm_first": true,
+      "t_norm_in": true,
+      "t_norm_in_group": false,
+      "t_norm_out": true,
+      "t_sin_random_shift": 0,
+      "t_sparse_attn_window": 400,
+      "t_sparse_cross_attn": false,
+      "t_sparse_self_attn": false,
+      "t_sparsity": 0.95,
+      "t_weight_decay": 0.05,
+      "t_weight_pos_embed": 1.0,
+      "time_stride": 2,
+      "wiener_iters": 0,
+      "wiener_residual": false
+    },
+    {
+      "audio_channels": 2,
+      "bottom_channels": 512,
+      "cac": true,
+      "channels": 48,
+      "channels_time": null,
+      "context": 1,
+      "context_enc": 0,
+      "dconv_comp": 8,
+      "dconv_depth": 2,
+      "dconv_init": 0.001,
+      "dconv_mode": 3,
+      "depth": 4,
+      "emb_scale": 10,
+      "emb_smooth": true,
+      "end_iters": 0,
+      "freq_emb": 0.2,
+      "growth": 2,
+      "kernel_size": 8,
+      "multi_freqs": [],
+      "multi_freqs_depth": 3,
+      "nfft": 4096,
+      "norm_groups": 4,
+      "norm_starts": 4,
+      "rescale": 0.1,
+      "rewrite": true,
+      "samplerate": 44100,
+      "segment": {
+        "__type__": "fraction",
+        "denominator": 5,
+        "numerator": 39
+      },
+      "sources": [
+        "drums",
+        "bass",
+        "other",
+        "vocals"
+      ],
+      "stride": 4,
+      "t_auto_sparsity": false,
+      "t_cape_augment": true,
+      "t_cape_glob_loc_scale": [
+        5000.0,
+        1.0,
+        1.4
+      ],
+      "t_cape_mean_normalize": true,
+      "t_cross_first": false,
+      "t_dropout": 0.02,
+      "t_emb": "sin",
+      "t_gelu": true,
+      "t_global_window": 100,
+      "t_group_norm": false,
+      "t_heads": 8,
+      "t_hidden_scale": 4.0,
+      "t_layer_scale": true,
+      "t_layers": 5,
+      "t_lr": null,
+      "t_mask_random_seed": 42,
+      "t_mask_type": "diag",
+      "t_max_period": 10000.0,
+      "t_max_positions": 10000,
+      "t_norm_first": true,
+      "t_norm_in": true,
+      "t_norm_in_group": false,
+      "t_norm_out": true,
+      "t_sin_random_shift": 0,
+      "t_sparse_attn_window": 400,
+      "t_sparse_cross_attn": false,
+      "t_sparse_self_attn": false,
+      "t_sparsity": 0.95,
+      "t_weight_decay": 0.05,
+      "t_weight_pos_embed": 1.0,
+      "time_stride": 2,
+      "wiener_iters": 0,
+      "wiener_residual": false
+    },
+    {
+      "audio_channels": 2,
+      "bottom_channels": 512,
+      "cac": true,
+      "channels": 48,
+      "channels_time": null,
+      "context": 1,
+      "context_enc": 0,
+      "dconv_comp": 8,
+      "dconv_depth": 2,
+      "dconv_init": 0.001,
+      "dconv_mode": 3,
+      "depth": 4,
+      "emb_scale": 10,
+      "emb_smooth": true,
+      "end_iters": 0,
+      "freq_emb": 0.2,
+      "growth": 2,
+      "kernel_size": 8,
+      "multi_freqs": [],
+      "multi_freqs_depth": 3,
+      "nfft": 4096,
+      "norm_groups": 4,
+      "norm_starts": 4,
+      "rescale": 0.1,
+      "rewrite": true,
+      "samplerate": 44100,
+      "segment": {
+        "__type__": "fraction",
+        "denominator": 5,
+        "numerator": 39
+      },
+      "sources": [
+        "drums",
+        "bass",
+        "other",
+        "vocals"
+      ],
+      "stride": 4,
+      "t_auto_sparsity": false,
+      "t_cape_augment": true,
+      "t_cape_glob_loc_scale": [
+        5000.0,
+        1.0,
+        1.4
+      ],
+      "t_cape_mean_normalize": true,
+      "t_cross_first": false,
+      "t_dropout": 0.02,
+      "t_emb": "sin",
+      "t_gelu": true,
+      "t_global_window": 100,
+      "t_group_norm": false,
+      "t_heads": 8,
+      "t_hidden_scale": 4.0,
+      "t_layer_scale": true,
+      "t_layers": 5,
+      "t_lr": null,
+      "t_mask_random_seed": 42,
+      "t_mask_type": "diag",
+      "t_max_period": 10000.0,
+      "t_max_positions": 10000,
+      "t_norm_first": true,
+      "t_norm_in": true,
+      "t_norm_in_group": false,
+      "t_norm_out": true,
+      "t_sin_random_shift": 0,
+      "t_sparse_attn_window": 400,
+      "t_sparse_cross_attn": false,
+      "t_sparse_self_attn": false,
+      "t_sparsity": 0.95,
+      "t_weight_decay": 0.05,
+      "t_weight_pos_embed": 1.0,
+      "time_stride": 2,
+      "wiener_iters": 0,
+      "wiener_residual": false
+    },
+    {
+      "audio_channels": 2,
+      "bottom_channels": 512,
+      "cac": true,
+      "channels": 48,
+      "channels_time": null,
+      "context": 1,
+      "context_enc": 0,
+      "dconv_comp": 8,
+      "dconv_depth": 2,
+      "dconv_init": 0.001,
+      "dconv_mode": 3,
+      "depth": 4,
+      "emb_scale": 10,
+      "emb_smooth": true,
+      "end_iters": 0,
+      "freq_emb": 0.2,
+      "growth": 2,
+      "kernel_size": 8,
+      "multi_freqs": [],
+      "multi_freqs_depth": 3,
+      "nfft": 4096,
+      "norm_groups": 4,
+      "norm_starts": 4,
+      "rescale": 0.1,
+      "rewrite": true,
+      "samplerate": 44100,
+      "segment": {
+        "__type__": "fraction",
+        "denominator": 5,
+        "numerator": 39
+      },
+      "sources": [
+        "drums",
+        "bass",
+        "other",
+        "vocals"
+      ],
+      "stride": 4,
+      "t_auto_sparsity": false,
+      "t_cape_augment": true,
+      "t_cape_glob_loc_scale": [
+        5000.0,
+        1.0,
+        1.4
+      ],
+      "t_cape_mean_normalize": true,
+      "t_cross_first": false,
+      "t_dropout": 0.02,
+      "t_emb": "sin",
+      "t_gelu": true,
+      "t_global_window": 100,
+      "t_group_norm": false,
+      "t_heads": 8,
+      "t_hidden_scale": 4.0,
+      "t_layer_scale": true,
+      "t_layers": 5,
+      "t_lr": null,
+      "t_mask_random_seed": 42,
+      "t_mask_type": "diag",
+      "t_max_period": 10000.0,
+      "t_max_positions": 10000,
+      "t_norm_first": true,
+      "t_norm_in": true,
+      "t_norm_in_group": false,
+      "t_norm_out": true,
+      "t_sin_random_shift": 0,
+      "t_sparse_attn_window": 400,
+      "t_sparse_cross_attn": false,
+      "t_sparse_self_attn": false,
+      "t_sparsity": 0.95,
+      "t_weight_decay": 0.05,
+      "t_weight_pos_embed": 1.0,
+      "time_stride": 2,
+      "wiener_iters": 0,
+      "wiener_residual": false
+    }
+  ],
+  "safetensors_sha256": "53f03b1ad4b4d211025a35da65460ba61a17547adf9c0544cad0ebcc8d7bbabb",
+  "source_artifacts": [
+    {
+      "checksum": "ba3fe64a",
+      "signature": "f7e0c4bc"
+    },
+    {
+      "checksum": "e57c48e6",
+      "signature": "d12395a8"
+    },
+    {
+      "checksum": "ef3bcb9c",
+      "signature": "92cfc3b6"
+    },
+    {
+      "checksum": "f3cf25b2",
+      "signature": "04573f0d"
+    }
+  ],
+  "sub_model_class": "HTDemucsMLX",
+  "verification_passed": true,
+  "weights": [
+    [
+      1.0,
+      0.0,
+      0.0,
+      0.0
+    ],
+    [
+      0.0,
+      1.0,
+      0.0,
+      0.0
+    ],
+    [
+      0.0,
+      0.0,
+      1.0,
+      0.0
+    ],
+    [
+      0.0,
+      0.0,
+      0.0,
+      1.0
+    ]
+  ]
+}

models/mel-roformer-instv7n/config.json

@@ -0,0 +1,57 @@
+{
+  "audio": {
+    "chunk_size": 485100,
+    "dim_f": 1024,
+    "dim_t": 1101,
+    "hop_length": 441,
+    "min_mean_abs": 0.0,
+    "n_fft": 2048,
+    "num_channels": 2,
+    "sample_rate": 44100
+  },
+  "inference": {
+    "batch_size": 1,
+    "dim_t": 1101,
+    "num_overlap": 2
+  },
+  "is_roformer": true,
+  "model": {
+    "attn_dropout": 0,
+    "depth": 6,
+    "dim": 384,
+    "dim_freqs_in": 1025,
+    "dim_head": 64,
+    "ff_dropout": 0,
+    "flash_attn": true,
+    "freq_transformer_depth": 1,
+    "heads": 8,
+    "mask_estimator_depth": 3,
+    "multi_stft_hop_size": 147,
+    "multi_stft_normalized": false,
+    "multi_stft_resolution_loss_weight": 1.0,
+    "multi_stft_resolutions_window_sizes": [
+      4096,
+      2048,
+      1024,
+      512,
+      256
+    ],
+    "num_bands": 60,
+    "num_stems": 1,
+    "sample_rate": 44100,
+    "stereo": true,
+    "stft_hop_length": 441,
+    "stft_n_fft": 2048,
+    "stft_normalized": false,
+    "stft_win_length": 2048,
+    "time_transformer_depth": 1
+  },
+  "training": {
+    "instruments": [
+      "Instrumental",
+      "Vocals"
+    ],
+    "target_instrument": "Instrumental",
+    "use_amp": true
+  }
+}

models/mel-roformer-instv7n/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74397de5395ed372059b971d6398106f60d92373ada69b62e986e52c9697a5ba
+size 912889223

models/mel-roformer-zfturbo-vocals-v1/LICENSE

@@ -0,0 +1,31 @@
+MIT License
+
+Copyright (c) 2026 Xocialize (MLX conversion and packaging)
+Copyright (c) 2023-2026 Roman Solovyev (ZFTurbo) — original Music-Source-Separation-Training
+                       repository and Mel-Band-RoFormer vocals_v1 release weights
+
+The original ZFTurbo Mel-Band-RoFormer vocals_v1 weights are distributed as a
+release asset of https://github.com/ZFTurbo/Music-Source-Separation-Training
+(release tag v1.0.0, asset model_vocals_mel_band_roformer_sdr_8.42.ckpt). The
+repository's LICENSE file (MIT) was added on 2024-11-04 in commit
+6149a22, approximately one year after the v1.0.0 release on 2023-11-06.
+This redistribution treats the weights as inheriting the repository's current
+MIT terms; see the model card for the full disclosure.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

models/mel-roformer-zfturbo-vocals-v1/config.json

@@ -0,0 +1,28 @@
+{
+  "_class": "MelRoFormerConfig",
+  "model_type": "mel_band_roformer",
+  "checkpoint_family": "zfturbo_vocals_v1",
+  "dim": 192,
+  "depth": 8,
+  "heads": 8,
+  "dim_head": 64,
+  "num_bands": 60,
+  "num_stems": 1,
+  "ff_mult": 4,
+  "mlp_expansion_factor": 4,
+  "mask_estimator_depth": 1,
+  "n_fft": 2048,
+  "hop_length": 512,
+  "win_length": 2048,
+  "sample_rate": 44100,
+  "chunk_size": 352800,
+  "num_overlap": 2,
+  "_dtype": "float16",
+  "_mlx_version": "0.31.0",
+  "_tool_version": "8380ab8",
+  "_source_input": "model_vocals_mel_band_roformer_sdr_8.42.ckpt",
+  "_source_sha256": "d9ce706b49cebf0af018590d8deb47ad5434987bf8f7bd3a87a4e5e8c30acb26",
+  "_source_repository": "https://github.com/ZFTurbo/Music-Source-Separation-Training",
+  "_source_release": "https://github.com/ZFTurbo/Music-Source-Separation-Training/releases/tag/v1.0.0",
+  "_source_license_at_conversion": "MIT (current MSS-Training repo LICENSE; LICENSE file added 2024-11-04 in commit 6149a22, post-dating the v1.0.0 release on 2023-11-06)"
+}

models/mel-roformer-zfturbo-vocals-v1/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef4aa052845a868cfaff93611477bd8f54d8081bc32f2742a9b3c738f0821191
+size 67402202

models/uvr-mdx-net-inst-hq3/config.json

@@ -0,0 +1,12 @@
+{
+  "bn": 8,
+  "compensate": 1.022,
+  "dim_c": 4,
+  "g": 48,
+  "l": 3,
+  "mdx_dim_f_set": 3072,
+  "mdx_dim_t_set": 8,
+  "mdx_n_fft_scale_set": 6144,
+  "num_blocks": 11,
+  "primary_stem": "Instrumental"
+}

models/uvr-mdx-net-inst-hq3/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:72c307239fbc415c9f3c1ef83fee38d19ac98b4b893f1a5afc4345296f48eea8
+size 66871394

validation/automatic-downloads.json

@@ -0,0 +1,298 @@
+{
+  "status": "passed",
+  "download_source": "local HTTP mirror of the publication bundle",
+  "public_hub_tested": false,
+  "fixture_sha256": "f5d0902f372a61ae2784239d0e097c9246c9d9b90be26a4a4805db8c96efad6d",
+  "demucs_shifts": 2,
+  "demucs_validation_seed": 0,
+  "roformer_overlap_windows": 2,
+  "source_sha256": {
+    "mlx_audio_separator/core.py": "fbac906edf6c125fbf8227050538b51e181a73820e92e7afb31969229cd7db04",
+    "mlx_audio_separator/model_hub.py": "96001b3d07b22b52dc727acbdfd37911d86e3d984ca7715fe3d88e1dd4118522",
+    "mlx_audio_separator/pretrained_models.json": "c1825eeda86b0739f206c5429e39b2a15e97bf78be12170512816391fe812c75",
+    "mlx_audio_separator/demucs_mlx/api.py": "64bcd2846826cb984753de005339712481e47c8d31367a6973f0071ba73a28cc",
+    "mlx_audio_separator/demucs_mlx/model_converter.py": "32c83cc1a13fe37b47e732a081be9d07dbb54a18081f6110dae70c318821f5da",
+    "mlx_audio_separator/separator/architectures/demucs_separator.py": "0dcf14eb910718d0ca3de41f8eafa31192e0a5a44ce7de929ac92845f866982f",
+    "scripts/catalog/validate_downloads.py": "7e38c0d1678235ad340d62f302c25588ae29fe98b526806089da68235da59f4d"
+  },
+  "models": [
+    {
+      "model": "bs-roformer-sw",
+      "legacy_name": "BS-Roformer-SW.ckpt",
+      "status": "passed",
+      "offline_output_bit_identical": true,
+      "stems": {
+        "bass": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.3247218430042267,
+          "raw_sha256": "e0308e6f3b8c4e82cb2b65e39e029b6f7d53c8527f58cf7de1aa1ca25ce2d10a"
+        },
+        "drums": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.00474020279943943,
+          "raw_sha256": "f526e6f84b02efc35d665085da92e37f2915617489e808b5d203940e12576271"
+        },
+        "other": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 2.648953886819072e-05,
+          "raw_sha256": "320f6cdbf3c128e5fdcd78a9e861ee900b9457000be542ea60084b13b037e1bb"
+        },
+        "vocals": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.27387580275535583,
+          "raw_sha256": "3e195538e564ce2ee672061f2019479fd2e4b3c1591f5dc687eaca30eab4f3b2"
+        },
+        "guitar": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 2.9653065212187357e-05,
+          "raw_sha256": "cae2ccee73471dcbc9c6bc86c6412faf28e7162b7688b72f4df71a4c4ffe2bdf"
+        },
+        "piano": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.3620210587978363,
+          "raw_sha256": "aec64b500b2a552fc18207066573679d0a474e55d61c0b66e4bcb0dbdd9548b0"
+        }
+      },
+      "conversion_dependencies_blocked": true,
+      "offline_network_blocked": true
+    },
+    {
+      "model": "mel-roformer-instv7n",
+      "legacy_name": "mel_band_roformer_instrumental_instv7n_gabox.ckpt",
+      "status": "passed",
+      "offline_output_bit_identical": true,
+      "stems": {
+        "instrumental": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.512724757194519,
+          "raw_sha256": "16eded8b94f233b67adb17bbdb2d4a794e85c4565829af8f66b39c2db6bc545a"
+        },
+        "vocals": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.29248708486557007,
+          "raw_sha256": "e809c0b8ac6cb5314e9e49e7cff4a0a0838da46669d1550d6e2bb2ad8cb2a856"
+        }
+      },
+      "conversion_dependencies_blocked": true,
+      "offline_network_blocked": true
+    },
+    {
+      "model": "uvr-mdx-net-inst-hq3",
+      "legacy_name": "UVR-MDX-NET-Inst_HQ_3.onnx",
+      "status": "passed",
+      "offline_output_bit_identical": true,
+      "stems": {
+        "vocals": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.32498443126678467,
+          "raw_sha256": "dea2889222b419d685a5d8c1fbe7717d8af4dd9234b9b8d3718df32323a34a19"
+        },
+        "instrumental": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.2586144208908081,
+          "raw_sha256": "0d15b30d669fe96a4214aa79839eb31029940d1e7b88cbd1182acf70726b6377"
+        }
+      },
+      "conversion_dependencies_blocked": true,
+      "offline_network_blocked": true
+    },
+    {
+      "model": "htdemucs",
+      "legacy_name": "htdemucs.yaml",
+      "status": "passed",
+      "offline_output_bit_identical": true,
+      "stems": {
+        "drums": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.0019488489488139749,
+          "raw_sha256": "2d1413830bf9faf27b534c5604e7df4d6b43a8864d701186ce00a32f8367d217"
+        },
+        "bass": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.36053600907325745,
+          "raw_sha256": "e1fb147bf64048a2a86909042a32693545c48c94fc42e00fbfd93f789947f772"
+        },
+        "other": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.36171454191207886,
+          "raw_sha256": "5da76923636e599e3919a75508cbff4e69de491f6964fd8f65b0a802988f2a3b"
+        },
+        "vocals": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.30794283747673035,
+          "raw_sha256": "402db8962619441481c78a98ae4266a3dbac05aea02573ff38a42d4e2414a02d"
+        }
+      },
+      "conversion_dependencies_blocked": true,
+      "offline_network_blocked": true
+    },
+    {
+      "model": "htdemucs_ft",
+      "legacy_name": "htdemucs_ft.yaml",
+      "status": "passed",
+      "offline_output_bit_identical": true,
+      "stems": {
+        "drums": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.0010993701871484518,
+          "raw_sha256": "021f3d9461374751c44e7af8e8c4252cbacf96e1a131586b352ea5026bb9d9c5"
+        },
+        "bass": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.358381986618042,
+          "raw_sha256": "c77c2fe756fe1d20ec8d223c87bedae54cf5a6ff562c5ab349b277852db13c0f"
+        },
+        "other": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.3595317602157593,
+          "raw_sha256": "6885bb26dc75bfe013154ebbe6035e2df501fcd650d6f60e8cb5b1ddbd5dada4"
+        },
+        "vocals": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.2946012616157532,
+          "raw_sha256": "d594e3bfdc97f1d6d96aab0ad9ee030f67520f713c759da53d9812ce2b67d791"
+        }
+      },
+      "conversion_dependencies_blocked": true,
+      "offline_network_blocked": true
+    },
+    {
+      "model": "htdemucs_6s",
+      "legacy_name": "htdemucs_6s.yaml",
+      "status": "passed",
+      "offline_output_bit_identical": true,
+      "stems": {
+        "drums": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.0005281093181110919,
+          "raw_sha256": "23386388e11b7d2e846eebf77a46d08ca55f075131679668fb8f6b80914bcf22"
+        },
+        "bass": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.3143438696861267,
+          "raw_sha256": "cf2e1c208da316f7ee660de9541220825d8fd8f59808f581f755abd21aa28ff8"
+        },
+        "other": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.0002693339192774147,
+          "raw_sha256": "65decc5bd490b7b393da0236101dc5f0ebc7b7ea1c3d3fd893e4cf00a2645a40"
+        },
+        "vocals": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.2923990786075592,
+          "raw_sha256": "bbaee6fdf2359590d64aec7e95313c1f75bdf9deae7441bc800f24c2db4cbcaf"
+        },
+        "guitar": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.00031150958966463804,
+          "raw_sha256": "b153959c43c377d755894527901b8457279228ae3699f64c0723dbb208b9c68e"
+        },
+        "piano": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.35046419501304626,
+          "raw_sha256": "4768791a4f50b9cd1d9a3c418a4141027f09f4c5ee0b209baed0b8575c9b2b9e"
+        }
+      },
+      "conversion_dependencies_blocked": true,
+      "offline_network_blocked": true
+    },
+    {
+      "model": "mel-roformer-zfturbo-vocals-v1",
+      "legacy_name": "mel-roformer-zfturbo-vocals-v1-mlx",
+      "status": "passed",
+      "offline_output_bit_identical": true,
+      "stems": {
+        "vocals": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.21046893298625946,
+          "raw_sha256": "0726c4b5da0b4745206f414de8b53a59aab7e72abe62dc7bfee11cf7a651383b"
+        },
+        "instrumental": {
+          "shape": [
+            617400,
+            2
+          ],
+          "peak": 0.5118863582611084,
+          "raw_sha256": "a835f80756717db407ffa42d6a7591c397edc7132ff8ef684aaf3e10f9f414cc"
+        }
+      },
+      "conversion_dependencies_blocked": true,
+      "offline_network_blocked": true
+    }
+  ]
+}

validation/bs-roformer-sw.json

@@ -0,0 +1,92 @@
+{
+  "model": "bs-roformer-sw",
+  "status": "passed",
+  "mode": "published",
+  "fixture_sha256": "f5d0902f372a61ae2784239d0e097c9246c9d9b90be26a4a4805db8c96efad6d",
+  "fixture_frames": 617400,
+  "sample_rate": 44100,
+  "stems": {
+    "bass": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.3247218430042267,
+      "finite": true
+    },
+    "drums": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.00474020279943943,
+      "finite": true
+    },
+    "other": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 2.648953886819072e-05,
+      "finite": true
+    },
+    "vocals": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.27387580275535583,
+      "finite": true
+    },
+    "guitar": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 2.9653065212187357e-05,
+      "finite": true
+    },
+    "piano": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.3620210587978363,
+      "finite": true
+    }
+  },
+  "seconds_load_separate_write": 1.4203401249833405,
+  "performance_measurement": false,
+  "network_disabled": true,
+  "conversion_dependencies_blocked": true,
+  "settings": {
+    "chunk_samples": 588800,
+    "overlap_windows": 2
+  },
+  "source_conversion_roundtrip": {
+    "bass": {
+      "max_abs_error": 0.0,
+      "bit_identical": true
+    },
+    "drums": {
+      "max_abs_error": 0.0,
+      "bit_identical": true
+    },
+    "other": {
+      "max_abs_error": 0.0,
+      "bit_identical": true
+    },
+    "vocals": {
+      "max_abs_error": 0.0,
+      "bit_identical": true
+    },
+    "guitar": {
+      "max_abs_error": 0.0,
+      "bit_identical": true
+    },
+    "piano": {
+      "max_abs_error": 0.0,
+      "bit_identical": true
+    }
+  }
+}

validation/htdemucs.json

@@ -0,0 +1,51 @@
+{
+  "model": "htdemucs",
+  "status": "passed",
+  "mode": "published",
+  "fixture_sha256": "f5d0902f372a61ae2784239d0e097c9246c9d9b90be26a4a4805db8c96efad6d",
+  "fixture_frames": 617400,
+  "sample_rate": 44100,
+  "stems": {
+    "drums": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.0019488489488139749,
+      "finite": true
+    },
+    "bass": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.36053600907325745,
+      "finite": true
+    },
+    "other": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.36171454191207886,
+      "finite": true
+    },
+    "vocals": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.30794283747673035,
+      "finite": true
+    }
+  },
+  "seconds_load_separate_write": 0.5797994589665905,
+  "performance_measurement": false,
+  "network_disabled": true,
+  "conversion_dependencies_blocked": true,
+  "settings": {
+    "overlap_fraction": 0.25,
+    "shifts": 2
+  },
+  "validation_seed": 0
+}

validation/htdemucs_6s.json

@@ -0,0 +1,67 @@
+{
+  "model": "htdemucs_6s",
+  "status": "passed",
+  "mode": "published",
+  "fixture_sha256": "f5d0902f372a61ae2784239d0e097c9246c9d9b90be26a4a4805db8c96efad6d",
+  "fixture_frames": 617400,
+  "sample_rate": 44100,
+  "stems": {
+    "drums": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.0005281093181110919,
+      "finite": true
+    },
+    "bass": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.3143438696861267,
+      "finite": true
+    },
+    "other": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.0002693339192774147,
+      "finite": true
+    },
+    "vocals": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.2923990786075592,
+      "finite": true
+    },
+    "guitar": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.00031150958966463804,
+      "finite": true
+    },
+    "piano": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.35046419501304626,
+      "finite": true
+    }
+  },
+  "seconds_load_separate_write": 0.864158249925822,
+  "performance_measurement": false,
+  "network_disabled": true,
+  "conversion_dependencies_blocked": true,
+  "settings": {
+    "overlap_fraction": 0.25,
+    "shifts": 2
+  },
+  "validation_seed": 0
+}

validation/htdemucs_ft.json

@@ -0,0 +1,51 @@
+{
+  "model": "htdemucs_ft",
+  "status": "passed",
+  "mode": "published",
+  "fixture_sha256": "f5d0902f372a61ae2784239d0e097c9246c9d9b90be26a4a4805db8c96efad6d",
+  "fixture_frames": 617400,
+  "sample_rate": 44100,
+  "stems": {
+    "drums": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.0010993701871484518,
+      "finite": true
+    },
+    "bass": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.358381986618042,
+      "finite": true
+    },
+    "other": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.3595317602157593,
+      "finite": true
+    },
+    "vocals": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.2946012616157532,
+      "finite": true
+    }
+  },
+  "seconds_load_separate_write": 1.800742125022225,
+  "performance_measurement": false,
+  "network_disabled": true,
+  "conversion_dependencies_blocked": true,
+  "settings": {
+    "overlap_fraction": 0.25,
+    "shifts": 2
+  },
+  "validation_seed": 0
+}

validation/mel-roformer-instv7n.json

@@ -0,0 +1,44 @@
+{
+  "model": "mel-roformer-instv7n",
+  "status": "passed",
+  "mode": "published",
+  "fixture_sha256": "f5d0902f372a61ae2784239d0e097c9246c9d9b90be26a4a4805db8c96efad6d",
+  "fixture_frames": 617400,
+  "sample_rate": 44100,
+  "stems": {
+    "instrumental": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.512724757194519,
+      "finite": true
+    },
+    "vocals": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.29248708486557007,
+      "finite": true
+    }
+  },
+  "seconds_load_separate_write": 1.1987386250402778,
+  "performance_measurement": false,
+  "network_disabled": true,
+  "conversion_dependencies_blocked": true,
+  "settings": {
+    "chunk_samples": 485100,
+    "overlap_windows": 2
+  },
+  "source_conversion_roundtrip": {
+    "instrumental": {
+      "max_abs_error": 0.0,
+      "bit_identical": true
+    },
+    "vocals": {
+      "max_abs_error": 0.0,
+      "bit_identical": true
+    }
+  }
+}

validation/mel-roformer-zfturbo-vocals-v1.json

@@ -0,0 +1,34 @@
+{
+  "model": "mel-roformer-zfturbo-vocals-v1",
+  "status": "passed",
+  "mode": "published",
+  "fixture_sha256": "f5d0902f372a61ae2784239d0e097c9246c9d9b90be26a4a4805db8c96efad6d",
+  "fixture_frames": 617400,
+  "sample_rate": 44100,
+  "stems": {
+    "vocals": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.21046893298625946,
+      "finite": true
+    },
+    "instrumental": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.5118863582611084,
+      "finite": true
+    }
+  },
+  "seconds_load_separate_write": 0.689663999946788,
+  "performance_measurement": false,
+  "network_disabled": true,
+  "conversion_dependencies_blocked": true,
+  "settings": {
+    "chunk_samples": 352800,
+    "overlap_windows": 2
+  }
+}

validation/uvr-mdx-net-inst-hq3.json

@@ -0,0 +1,44 @@
+{
+  "model": "uvr-mdx-net-inst-hq3",
+  "status": "passed",
+  "mode": "published",
+  "fixture_sha256": "f5d0902f372a61ae2784239d0e097c9246c9d9b90be26a4a4805db8c96efad6d",
+  "fixture_frames": 617400,
+  "sample_rate": 44100,
+  "stems": {
+    "vocals": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.32498443126678467,
+      "finite": true
+    },
+    "instrumental": {
+      "shape": [
+        617400,
+        2
+      ],
+      "peak": 0.2586144208908081,
+      "finite": true
+    }
+  },
+  "seconds_load_separate_write": 0.7514086659066379,
+  "performance_measurement": false,
+  "network_disabled": true,
+  "conversion_dependencies_blocked": true,
+  "settings": {
+    "overlap_fraction": 0.25,
+    "segment_size": 256
+  },
+  "source_conversion_roundtrip": {
+    "vocals": {
+      "max_abs_error": 0.0,
+      "bit_identical": true
+    },
+    "instrumental": {
+      "max_abs_error": 0.0,
+      "bit_identical": true
+    }
+  }
+}