ddalcu/Qwen-Image-2.1-MLX-Serve-8bit

Qwen-Image-2.1 MLX-Serve 8-bit

8-bit pack of Qwen/Qwen-Image-2.1 for mlx-serve: 17.6 GB, for 32 GB Macs.

Not released yet. These packs load only on the mlx-serve branch feat/qwen-image-2.1. No released mlx-serve or MLX Core build can run them. This notice goes away when the PR ships.

What is in it

The checkpoint's own diffusers layout and key names, with the DiT block linears and the text-encoder layer linears affine-quantized to 8-bit (group 64). Kept dense: the VAE (f32), embed_tokens, norms, and the DiT's small or shared linears. Dropped: the Qwen3-VL vision tower and lm_head (text-to-image only) and the VAE's per-frame time_convs. Built by tests/convert_qwen_image21_weights.py --preset 32gb.

Measured (M1 Pro, 32 GB)

Pack	Size	Steps	Wall clock incl. load	Peak memory
8-bit	1024x1024	40	985 s (~23 s/step)	12.95 GB
4-bit	1024x1024	3	87 s	9.55 GB
4-bit	512x512	20	118 s	-

On a Mac the full set would crowd, mlx-serve loads the text encoder per request and frees it before the denoise, so the resident set is the DiT and VAE.

Run it (from the branch)

git clone -b feat/qwen-image-2.1 https://github.com/ddalcu/mlx-serve && cd mlx-serve
./scripts/fetch-zig.sh && ./scripts/build-mlx.sh && .zig-toolchain/zig build -Doptimize=ReleaseFast
./zig-out/bin/mlx-serve pull ddalcu/Qwen-Image-2.1-MLX-Serve-8bit
./zig-out/bin/mlx-serve serve
curl localhost:11234/v1/images/generations -H 'Content-Type: application/json' \
  -d '{"model":"ddalcu/Qwen-Image-2.1-MLX-Serve-8bit","prompt":"a red fox in fresh snow","size":"1024x1024"}'

40 steps when steps is omitted. guidance_scale above 1 with a negative_prompt runs real CFG (two forwards per step). image + strength does image-to-image.

Apache-2.0, same as the base model.