---
base_model:
- Tongyi-MAI/Z-Image-Turbo
pipeline_tag: text-to-image
license: other
tags:
- quantized
- mxfp4
- autoround
- diffusion
- text-to-image
- autoquant-agent
---

# Z-Image-Turbo-MXFP4-RTN-AutoRound

## Model Details

This is a **MXFP4** (4-bit micro-scaling) quantization of [Tongyi-MAI/Z-Image-Turbo](https://huggingface.co/Tongyi-MAI/Z-Image-Turbo), a 6B S3-DiT distilled text-to-image model. Generated by [AutoRound](https://github.com/intel/auto-round) with RTN (round-to-nearest, iters=0).

- **Base model:** [Tongyi-MAI/Z-Image-Turbo](https://huggingface.co/Tongyi-MAI/Z-Image-Turbo)
- **Quantization:** MXFP4 (W4A4), group_size=32
- **Method:** AutoRound RTN
- **Model size:** ~11 GB (vs 31 GB bf16)

## Quantization Details

- **Scheme:** MXFP4 (data_type=mx_fp, bits=4, act_bits=4)
- **Group size:** 32
- **Export format:** auto_round (vllm-omni compatible)
- **Calibration:** coco2014, 8 steps, guidance 0.0
- **Ignored layers:** adaLN_modulation (kept full precision)

## Evaluation

Evaluated with vllm-omni diffusion harness (8 steps, guidance 0.0, 1024×1024, seed 42).

| Benchmark | BF16 Baseline | MXFP4 Quantized |
|---|---|---|
| DrawBench CLIP | 31.73 | 31.66 |
| DrawBench CLIP-IQA | 70.57 | 68.52 |
| DrawBench ImageReward | 1.00 | 0.91 |
| GenEval | 0.757 | 0.741 |

MXFP4 quantization is nearly lossless vs the BF16 baseline (GenEval 0.741 vs 0.757, CLIP 31.66 vs 31.73).

## Usage

```python
from vllm_omni.entrypoints.omni import Omni
from vllm_omni.inputs.data import OmniDiffusionSamplingParams

omni = Omni(model="INCModel3/Z-Image-Turbo-MXFP4-RTN-AutoRound", mode="text-to-image")
params = OmniDiffusionSamplingParams(
    height=1024, width=1024, seed=42,
    guidance_scale=0.0, num_inference_steps=8, num_outputs_per_prompt=1,
)
out = omni.generate("a red bench in a park", sampling_params_list=[params])
```

## License

Please follow the license of the original model [Tongyi-MAI/Z-Image-Turbo](https://huggingface.co/Tongyi-MAI/Z-Image-Turbo).

_Produced with [autoquant-agent](https://github.com/) — agent-driven quantize + evaluate + self-heal._