Instructions to use mlx-community/ACE-Step-1.5-text-encoder-bf16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use mlx-community/ACE-Step-1.5-text-encoder-bf16 with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download mlx-community/ACE-Step-1.5-text-encoder-bf16 --local-dir ACE-Step-1.5-text-encoder-bf16
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Download NOTICE from mlx-community/ACE-Step-1.5-text-encoder-bf16: direct link, hf CLI and curl.
- Browser
- Download file 1.07 kB
-
https://huggingface.co/mlx-community/ACE-Step-1.5-text-encoder-bf16/resolve/main/NOTICE
- Command line
-
hf download hf://mlx-community/ACE-Step-1.5-text-encoder-bf16/NOTICE
-
curl -L -o NOTICE https://huggingface.co/mlx-community/ACE-Step-1.5-text-encoder-bf16/resolve/main/NOTICE
1.07 kB
| ACE-Step 1.5 — Qwen3-Embedding-0.6B (text encoder, Apache-2.0, Qwen) — by ACE Studio and StepFun (https://huggingface.co/ACE-Step/Ace-Step1.5, ACE-Step/Ace-Step1.5@19671f4). | |
| This bundle is an MLX-layout conversion of the released checkpoint (key prefix / conv layout / weight_norm folding); | |
| the model is otherwise unmodified. Code licence: MIT (ace-step/ACE-Step-1.5; LICENSE is that repository's text). | |
| Weights: the Hugging Face repository has no LICENSE file; its card tag and README say MIT, and the GitHub LICENSE | |
| (Copyright (c) 2026 ACEStep) is the controlling text. Footnote 1: the HF repo carries no LICENSE file — the grant | |
| is the card tag + README + the GitHub LICENSE. Footnote 2: the Oobleck VAE (ACE-Step/ace-step-v1.5-1d-vae- | |
| stable-audio-format, MIT) is architecturally Stable Audio Open's VAE retargeted to 48 kHz; whether it was | |
| initialised from Stability's community-licensed weights is stated nowhere — ACE-Step asserts MIT and ships it. | |
| Bases: Qwen3-Embedding-0.6B and the planner's Qwen3 base are Apache-2.0 (Qwen); the fine-tune is ACE-Step's. | |