ACE-Step 1.5 Turbo MLX for AI2Apps

Converted from ACE-Step/Ace-Step1.5 revision 19671f406d603126926c1b7e2adc169acbcade22. Includes Turbo DiT, VAE, Qwen3-Embedding-0.6B and 1.7B planning LM. No quantization was added.

Modified tensors: convolution layouts transposed for MLX, LM keys prefixed with model., silence latent converted from a weights-only PyTorch tensor to NumPy. Original component precision is preserved. Conversion script is experiments/music_mlx/convert_ace.py in the AI2Apps Local project. Runtime uses the pinned mlx-audio commit in provenance.json.

ACE-Step is MIT licensed. Qwen3 components are subject to Apache-2.0. See the included component license notices. Package model code and checkpoint distributions are published separately; no runtime downloads or PyTorch fallback are needed.

Downloads last month
42
Safetensors
Model size
2B params
Tensor type
BF16
·
MLX
Hardware compatibility
Log In to add your hardware

Quantized

Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Avdpro/ACE-Step-1.5-Turbo-MLX

Finetuned
(15)
this model