---
title: HSWQ-fp8-SDXL
canonical_url: "https://www.modelscope.cn/models/ApacheOne/HSWQ-fp8-SDXL"
md_url: "https://www.modelscope.cn/models/ApacheOne/HSWQ-fp8-SDXL.md"
repository: ApacheOne/HSWQ-fp8-SDXL
last_updated: 2026-04-21
license: agpl-3.0
model_type:
  - custom
parameters: 12.1B
tensor_type:
  - I64
  - F16
  - F8_E4M3
  - F32
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
downloads: 66
stars: 0
---

# HSWQ-fp8-SDXL

> HSWQ-fp8-SDXL - ApacheOne 在 ModelScope 开源的模型。Creator: https://civitai.com/user/jice

ApacheOne/HSWQ-fp8-SDXL 是 ModelScope 魔搭社区上的 12.1B 参数机器学习模型，采用 agpl-3.0 许可。

- **Repository**: ApacheOne/HSWQ-fp8-SDXL
- **License**: agpl-3.0
- **Parameters**: 12.1B
- **Downloads**: 66
- **Stars**: 0
- **Last updated**: 2026-04-21

Source: https://www.modelscope.cn/models/ApacheOne/HSWQ-fp8-SDXL

---

# Model info

Creator: [https://civitai.com/user/jice](https://civitai.com/user/jice)

[https://civitai.com/models/383364?modelVersionId=471056](https://civitai.com/models/383364?modelVersionId=471056)

`creapromptLightning_creapromtHypersdxlV1_r32_r0.1_HSWQ_fp8e4m3.safetensors` full model

`creapromptLightning_creapromtHypersdxlV1_r32_r0.1_HSWQ_fp8e4m3_unetonly.safetensors` unet only can use my [NVFP4 clips](https://huggingface.co/ApacheOne/sdxl_text_encoders-NVFP4) with taesd vae [https://github.com/madebyollin/taesd](https://github.com/madebyollin/taesd)

[https://civitai.com/models/383364?modelVersionId=505350](https://civitai.com/models/383364?modelVersionId=505350)

`creapromptLightning_creapromptHyperCFGV2_r32_r0.1_HSWQ_fp8e4m3.safetensors` full model 

`creapromptLightning_creapromptHyperCFGV2_r32_r0.1_HSWQ_fp8e4m3_unetonly.safetensors` unet only can use my [NVFP4 clips](https://huggingface.co/ApacheOne/sdxl_text_encoders-NVFP4) with taesd vae [https://github.com/madebyollin/taesd](https://github.com/madebyollin/taesd)

# Hybrid-Sensitivity-Weighted-Quantization (HSWQ)

<p align="center">
  <img src="https://raw.githubusercontent.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization/main/icon.png" width="128">
</p>

High-fidelity FP8 quantization for diffusion models (SDXL). HSWQ uses **sensitivity** and **importance** analysis instead of naive uniform cast, and offers two modes: standard-compatible (V1) and high-performance scaled (V2).

**Technical details:** [md/HSWQ_ Hybrid Sensitivity Weighted Quantization.md](https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization/blob/main/md/HSWQ_%20Hybrid%20Sensitivity%20Weighted%20Quantization.md)

**How to quantize:** [md/HSWQ_ How to quantize SDXL.md](https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization/blob/main/md/How%20to%20quantize%20SDXL.md)

**SDXL Benchmark Test Results:** [md/SDXL Benchmark Test Results.md](https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization/blob/main/test/benchmark_test.md)





# Credit & Special Acknowledgement

[https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization](https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization)

[https://github.com/tritant/ComfyUI_Kitchen_nvfp4_Converter](https://github.com/tritant/ComfyUI_Kitchen_nvfp4_Converter)

[https://github.com/NVIDIA/Model-Optimizer](https://github.com/NVIDIA/Model-Optimizer)

We extend our deepest respect and gratitude to the **Nunchaku Team** for their groundbreaking work on SVDQ quantization and for sharing their models with the community. This collection relies heavily on their research and original implementation.
- **Original Repository:** [nunchaku-tech/nunchaku-sdxl](https://huggingface.co/nunchaku-tech/nunchaku-sdxl)