Instructions to use ionden/taew2.1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use ionden/taew2.1 with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir taew2.1 ionden/taew2.1
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
| license: mit | |
| library_name: mlx | |
| tags: | |
| - tiny-autoencoder | |
| - taehv | |
| - taew2.1 | |
| - qwen-image | |
| - wan2.1 | |
| - mlx | |
| # taew2.1 (re-host) | |
| A faithful re-host of madebyollin's **taew2.1** — the Tiny AutoEncoder for the Wan 2.1 VAE | |
| (used by Qwen-Image / Qwen-Image-Edit) — for use by | |
| [`mlx-taef`](https://github.com/IonDen/mlx-taef) as a fast, low-memory preview decoder. | |
| ## Provenance | |
| - **Canonical source:** [`madebyollin/taehv`](https://github.com/madebyollin/taehv) @ commit | |
| `a1c8e6a2ba77b91f284ef98935ec5bd21a41d786`, file `safetensors/taew2_1.safetensors`. | |
| - **sha256:** `04766eac0221b5390b985ae3fdcca652cbb4b1e8b82b28ea7ff89dfad1b1a93f` — byte-identical to | |
| the canonical GitHub file. fp16, 128 tensors, one combined file holding both encoder and decoder. | |
| - **License:** MIT (unchanged from upstream). | |
| This repo exists only because the canonical weights are published on GitHub (madebyollin hosts no | |
| Hugging Face copy), and re-uploaded community mirrors were found to differ from the canonical | |
| weights. `mlx-taef` pins this repo by revision + sha256 so the bytes are verified at download time. | |