Downloads · 30 days
581
24% of all-time downloads
stuqiu/nunchaku-qwen-image-2512
nunchaku-qwen-image-2512 is a text-to-image model from stuqiu. Use it when you need an image from a text prompt. It is set up for diffusers. The card lists the license as apache-2.0.
4-bit SVDQuant of Qwen/Qwen-Image-2512 for the nunchaku inference engine. Same file layout and usage as the official nunchaku-tech/nunchaku-qwen-image release — load it with stock frompretrained, no extra code.
Downloads · 30 days
581
24% of all-time downloads
All-time downloads
2.4K
Public
Repo size
25 GB
Likes
3
Public
Click a slice to open those files.
.safetensors25 GB · 100%
From the Hugging Face model README
4-bit SVDQuant of Qwen/Qwen-Image-2512 for the
nunchaku inference engine. Same file layout and usage
as the official nunchaku-tech/nunchaku-qwen-image
release — load it with stock from_pretrained, no extra code.
| Data Type | Rank | Model Name | Comment |
|---|---|---|---|
| NVFP4 | r32 | svdq-fp4_r32-qwen-image-2512.safetensors | ~12 GB · smaller / faster · default |
| r128 | svdq-fp4_r128-qwen-image-2512.safetensors | ~13 GB · higher fidelity (finer detail & text edges) |
Both checkpoints are W4A4 NVFP4 and differ only in the SVDQuant low-rank branch size. NVFP4 needs a Blackwell GPU (RTX 50-series sm_120 / GB10 sm_121); INT4 builds are not provided for this finetune yet.
Identical to the official nunchaku Qwen-Image example — two lines changed:
import torch
from diffusers import QwenImagePipeline
from nunchaku.models.transformers.transformer_qwenimage import NunchakuQwenImageTransformer2DModel
from nunchaku.utils import get_precision
rank = 32 # 32 (smaller/faster) or 128 (higher fidelity)
transformer = NunchakuQwenImageTransformer2DModel.from_pretrained(
f"stuqiu/nunchaku-qwen-image-2512/svdq-{get_precision()}_r{rank}-qwen-image-2512.safetensors"
)
pipe = QwenImagePipeline.from_pretrained(
"Qwen/Qwen-Image-2512", transformer=transformer, torch_dtype=torch.bfloat16
).to("cuda")
image = pipe(
prompt="夜晚的中式茶馆门口,门匾上写着「天下为公」四个金色大字",
num_inference_steps=50, true_cfg_scale=4.0,
).images[0]
image.save("out.png")
1024×1024, 50 steps, stock pipeline:
| Hardware | Precision | per-step | 50-step image | GPU mem after load |
|---|---|---|---|---|
| RTX 5090 (sm_120) | this model (W4A4) | 238 ms (~5.0 it/s) | 11.9 s | 26.8 GiB |
| RTX 5090 (sm_120) | BF16 | n/a — model does not fit in 32 GB, requires CPU offloading (much slower) | ||
| DGX Spark GB10 (sm_121) | this model (W4A4) | 1.06 s | ~55 s | — |
| DGX Spark GB10 (sm_121) | BF16 | 1.80 s | ~90 s | — |
Pipeline load 4.5 s, warmup step 2.6 s (RTX 5090).
Generated on RTX 5090 with the rank-32 checkpoint (50 steps, 1024×1024); see the rank 32 vs rank 128 comparison below:
![]() | ![]() |
8k high quality portrait of a beautiful woman, sharp focus, detailed face | 8k photorealistic landscape, golden hour, mountains and lake |

夜晚的中式茶馆门口,门匾上写着「天下为公」四个金色大字
| rank 32 | rank 128 |
|---|---|
![]() | ![]() |
夜晚的中式茶馆门口,门匾上写着「天下为公」四个金色大字
| rank 32 | rank 128 |
|---|---|
![]() | ![]() |
A cat sitting on a chair
| rank 32 | rank 128 |
|---|---|
![]() | ![]() |
A 20-year-old East Asian girl with delicate features and large bright brown eyes, naturally wavy long hair, fair skin and light makeup, in a cute bright-colored dress, standing indoors at an anime convention surrounded by banners and posters — casual iPhone-snapshot look.
deepcompressor SVDQuant calibration
(64 samples, smoothing grid 10, rank-32 low-rank branches) for attention/MLP, official symmetric
AWQ recipe (scale = absmax/7, zero ≡ 7) for modulation, plus the super-channel fold above.