This repository contains a full Diffusers pipeline for baidu/ERNIE-Image-Turbo with the transformer replaced by a Nunchaku Lite quantized transformer.

The tokenizer, text encoder, prompt enhancer, scheduler, and VAE are inherited from the original ERNIE-Image-Turbo release. Only the Diffusion Transformer component is replaced.

Quantization

The transformer uses a compact Nunchaku Lite quantization config stored in transformer/config.json.

Op	Precision	Group size	Rank	Targets
`svdq_w4a4`	`fp4` / NVFP4 runtime	16	32	252
`awq_w4a16`	`int4`	64	-	6

Requirements

This checkpoint requires a Diffusers version that includes NunchakuLiteQuantizationConfig and the Nunchaku Lite loader path.

bash

pip install git+https://github.com/huggingface/diffusers
pip install kernels

Use a CUDA-capable PyTorch environment. The quantized transformer uses kernels from the Hugging Face kernels package.

Usage

python

import torch
from diffusers import ErnieImagePipeline

pipe = ErnieImagePipeline.from_pretrained(
    "rootonchair/ERNIE-Image-Turbo-nunchaku-lite-nvfp4",
    torch_dtype=torch.bfloat16,
).to("cuda")

image = pipe(
    prompt="A modern red armchair in a quiet studio, soft window light, realistic product photography",
    height=1024,
    width=1024,
    num_inference_steps=8,
    guidance_scale=1.0,
    use_pe=False,
).images[0]

image.save("ernie-image-turbo-nunchaku-lite-nvfp4.png")

Recommended ERNIE-Image-Turbo settings:

num_inference_steps=8
guidance_scale=1.0
use_pe=False for direct prompt usage, or use_pe=True to enable the inherited prompt enhancer

Provenance

Base pipeline: baidu/ERNIE-Image-Turbo
Source quantized checkpoint: rootonchair/ERNIE-Image-Turbo-nunchaku-lite

Limitations

This is an experimental quantized checkpoint.
Only the transformer is quantized; the text encoder, prompt enhancer, scheduler, and VAE are inherited from the base model.
save_pretrained() serialization for Nunchaku Lite models is not supported yet.
Output quality and performance may differ from the original BF16 transformer.

License

This repository follows the Apache-2.0 license from the base ERNIE-Image-Turbo release.