LocalMuseAI/mlx-moss-image-sft-dmd-4bit

MOSS-Image 8-Step — MLX INT4 for LocalMuse

On-device package for the LocalMuse iOS app (profile mossImage5B4Bit).

• Model: MOSS-Image-SFT by the OpenMOSS Team, revision fc5de0d56fc910cbca6f6625a70f5f582b26deb1 (Apache-2.0): 5B single-stream DiT, Qwen3.5 text encoder, FLUX.2 small decoder.
• 8-step distillation: lora_latest.safetensors from lodestones/MOSS-Image-SFT-8-steps-dmd, revision f95df83b5e90ae91ea412434b3f9b7f1c8c6314f (Apache-2.0), a work-in-progress DMD distillation of the 50-step, CFG 3 teacher.

This is a format conversion, not a new fine-tune. Every source file is authenticated by source.json. Preserve LICENSE, NOTICE and the attribution.

Contents

File	Content
text_encoder/model.safetensors	Qwen3.5 text model (visual tower dropped), all linears and the embedding table affine INT4, group 64
transformer/model.safetensors	MOSS-Image DiT: attention/FFN INT4 group 64; embedders, time embedding, modulation, norms and output projection BF16
transformer/adapter.safetensors	the DMD adapter as BF16 rank-128 factors, applied unmerged at strength 0.7
vae.safetensors	FP32 FLUX.2 small decoder and latent BatchNorm statistics
tokenizer/	release tokenizer
runtime_config.json, manifest.json	runtime contract and SHA-256 manifest

The adapter must not be merged into the weights: its delta is smaller than BF16 rounding of the merged matrices, so merging (and quantizing a merge) loses the distillation. LocalMuse adds 0.7 * up(down(x)) after each quantized block projection, as the authors' demo Space does.

Inference contract

Text to Image, batch one, the release MOSSImageScheduler (flow-matching Euler, shift 5, sigmas snapped to the 1000 training sigmas), 8 steps (4–8 accepted), no classifier-free guidance and no negative prompt. Canvases: 1024 × 1024 and 896 × 1152. Prompts use the release Qwen3.5 chat template with thinking enabled; natural language, tags and MOSS JSON layouts are accepted.

INT4 is not bit-identical to BF16: with the same seed the composition, subject and lighting match while clothing, hair and brightness details can differ. Conversion, independent references and measurements are in the LocalMuse repository under Tools/MOSSImage.

Samples

Native LocalMuse Swift runtime on an M1 Pro, 8 steps, seed 1729 and 2026.

1024 × 1024	896 × 1152