Modelpublic

wabibito/Onyx-Z-Image-3bit

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face

0likes4downloads

Model Card

Onyx Z Image — 3-bit MLX

A 3-bit MLX quantization of the official `Tongyi-MAI/Z-Image`, built for on-device inference in the Onyx app with a documented, reproducible pipeline.

—Source: official Tongyi-MAI/Z-Image (diffusers), quantized in-house.
—Quantization: 3-bit, group 64, MLX. Only transformer + text-encoder Linear leaves are quantized; norms/embeddings/convs and the whole VAE stay fp16. Components: transformer/, text_encoder/ (Qwen3), vae/ (16-ch FLUX-family), tokenizer/, scheduler/.
—Inference: ~22-28 steps + real guidance (undistilled). Built as MIXED 3/4-bit: quant-sensitive layers (embedders, final layer, adaLN modulation, attention) stay 4-bit; the FFN bulk is 3-bit. Uniform 3-bit mottles the undistilled DiT; this recovers a clean image at near-3-bit size.
—Verified: reloaded through the Onyx engine and render-checked before publish.

Quantized derivative for efficient on-device use; all credit for the model to Tongyi-MAI (Alibaba).