Downloads · 30 days
5K
99% of all-time downloads
Qwen-Image-2.1-nunchaku-lite-int4
BlazeMCworld/Qwen-Image-2.1-nunchaku-lite-int4
Qwen-Image-2.1-nunchaku-lite-int4 is a machine learning model from BlazeMCworld. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for diffusers. The card lists the license as other.
All-time downloads
5K
Public
Parameters
3.8B
4.1 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors4.1 GB · 100%
Parameter types
How the weights are stored.
I83.5B · 91%
Base models
Model card
From the Hugging Face model README
<img src="image.webp" width="250"/>
Based on Qwen-Image-2.1. Only includes the transformer component.
At the point of uploading the latest diffusers==0.40.0 does NOT include any required changes.
Loading the pre-quantized checkpoint works using a source install from main, quantizing needs this PR as well.
Config used:
QwenImage21Transformer2DModel.from_pretrained(
"./qwen-image-2.1/transformer", torch_dtype=torch.bfloat16,
quantization_config=NunchakuLiteQuantizationConfig(
svdq_w4a4={
"precision": "int4",
"group_size": 64,
"rank": 32,
},
pre_quantized=False,
),
device_map="cuda"
).save_pretrained("./result")
The pull request source was also slightly adjusted to use svd_lowrank instead of torch.linalg.svd during quantizing.
- Library
- diffusers
- License
- other
- Created
- Sep 23, 2026
- Updated
- Sep 23, 2026