| --- |
| license: other |
| tags: |
| - text-to-image |
| - sdxl |
| - nunchaku |
| - svdq |
| - quantized |
| - int8 |
| - illustrious |
| - realvisxl |
| - photorealistic |
| - comfyui |
| - controlnet |
| - anime |
| - faipl-1.0-sd |
| - creativeml-openrail-m |
| library_name: nunchaku |
| --- |
| |
|
|
| # 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 ConvRot INT8 quantization for diffusion models (SDXL). HSWQ uses **sensitivity** and **importance** analysis instead of naive uniform cast. This is highly useful for users who need to strictly manage their VRAM resources while maintaining maximum image quality. |
|
|
| ComfyUI-compatible `int8_tensorwise` pack with **FULL ConvRot** on remaining Linear/Conv2d after DualMonitor + V4 weighted-histogram FP16 protection under a fixed **300 MiB** budget. Keep ratio is **0** (r0); critical layers stay FP16 via automatic analysis, not a keep-ratio percentage. SDXL pack scripts: `quantize_sdxl_hswq_v3.1.py` (HSWQ) and `native_convert_int8_sdxl.py` (native). |
|
|
| **Technical details:** [https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization](https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization) |
|
|
| **How to quantize (SDXL ConvRot INT8):** [md/How to quantize SDXL.md](https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization/blob/main/md/How%20to%20quantize%20SDXL.md) |
|
|
| **ComfyUI Loader for ConvRot INT8 / INT8:** To load these INT8 models in ComfyUI, please use the unofficial loader node: [ComfyUI-nunchaku-unofficial-loader](https://github.com/ussoewwin/ComfyUI-nunchaku-unofficial-loader) |
|
|
| **Post-quantize fidelity bench (integrated, default ON):** After save, `quantize_sdxl_hswq_v3.1.py` and `native_convert_int8_sdxl.py` **clear parent VRAM**, then automatically run `benchmark/int8bench_sdxl.py` with `--fp16` = the FP16 input, `--int8` = the saved pack, and a fixed `--prompt` / `--seed` (not inventable parent CLI overrides). Pass `--no-bench` to skip. Standalone re-runs use the same `int8bench_sdxl.py` command shape as in the How-to. |
|
|
| **SDXL ConvRot INT8 Benchmark Test Results (published tables):** [test/benchmark_sdxl_int8.md](https://github.com/ussoewwin/Hybrid-Sensitivity-Weighted-Quantization/blob/main/test/benchmark_sdxl_int8.md) |
|
|
| --- |
|
|
| ## Benchmark (Reference) |
|
|
| | Model | SSIM (Avg) | File size | Compatibility | |
| | :--- | :--- | :--- | :--- | |
| | Original FP16 | 1.0000 | 100% | High | |
| | Naive INT8 | 0.95-0.97 | 50% | High | |
| | **HSWQ ConvRot INT8** | **0.94-0.98** | 68% (FP16 mixed) | **High** (ComfyUI INT8) | |
|
|
| --- |
|
|
| ## 📦 Available Models |
|
|
| | Filename | Base Model | Version | License | |
| | :--- | :--- | :--- | :--- | |
| | `JANKUTrainedChenkinNoobai_v777_hswq_r32_1off_convrot_int8.safetensors` | [JANKU Trained Chenkin & Noobai-Rouwei (Illustrious-XL)](https://civitai.com/models/1277670/janku-trained-chenkin-and-noobai-rouwei-illustrious-xl) | v777 | Fair AI Public License 1.0-SD | |
| | `bluePencilXL_v031_hswq_r32_1off_convrot_int8.safetensors` | [blue_pencil-XL](https://civitai.com/models/119012) | v0.3.1 | CreativeML Open RAIL++-M | |
| | `epicrealismXL_pureFix_hswq_r32_1off_convrot_int8.safetensors` | [epiCRealism XL](https://civitai.com/models/277058) | pureFix | CreativeML Open RAIL++-M | |
| | `koronemixIllustrious_v70_sci_1on_covrot_int8.safetensors` | koronemixIllustrious | v70 | Fair AI Public License 1.0-SD | |
| | `koronemixVpred_v20_sci_1off_convrot_int8.safetensors` | koronemixVpred | v2.0 | CreativeML Open RAIL++-M | |
| | `novaAnimeXL_ilV190_hswq_r32_1on_convrot_int8.safetensors` | Nova Anime XL | ilV190 | Fair AI Public License 1.0-SD | |
| | `novaAsianXL_illustriousV70_hswq_r32_1off_convrot_int8.safetensors` | Nova Asian XL | v7.0 | Fair AI Public License 1.0-SD | |
| | `oneObsession_v23_hswq_r32_1off_convrot_int8.safetensors` | [OneObsession](https://civitai.com/models/691062) | v23 | CreativeML Open RAIL++-M | |
| | `prefectIllustriousXL_v8_hswq_r32_1on_convrot_int8.safetensors` | Prefect Illustrious XL | v8 | Fair AI Public License 1.0-SD | |
| | `realvisxlV30_v30TurboBakedvae_hswqr32_r32_1on_convrot_int8_.safetensors` | [RealVisXL V3.0 (Turbo)](https://civitai.com/models/139562?modelVersionId=361593) | v3.0 Turbo | CreativeML Open RAIL++-M | |
| | `realvisxlV50_v40Bakedvae_hswq_r32_ioff_convrot_int8.safetensors` | [RealVisXL V5.0 (Lightning)](https://civitai.com/models/139562/realvisxl-v50) | v4.0 BakedVAE | CreativeML Open RAIL++-M | |
| | `realvisxlV50_v50Bakedvae_hswq_r32_1on_covrot_int8.safetensors` | [RealVisXL V5.0 (Lightning)](https://civitai.com/models/139562/realvisxl-v50) | v5.0 BakedVAE | CreativeML Open RAIL++-M | |
| | `uwazumimixILL_v50_hswq_r32_1on_convrot_int8.safetensors` | UwazumiMix | v5.0 | Fair AI Public License 1.0-SD | |
| | `waiIllustriousSDXL_v170hswq_r32_1off_convrot_int8.safetensors` | [Illustrious-XL v1.7 (WAI-illustrious-SDXL)](https://civitai.com/models/827184/wai-illustrious-sdxl) | v17.0 (HF weight) | Fair AI Public License 1.0-SD | |
| | `waiREALCN_v150_hswq_r32_1on_convrot_int8.safetensors` | WAI-REAL_CN | v15.0 | Fair AI Public License 1.0-SD | |
| | `waiREALISM_v10_hswq_r32_1on_convrot_int8.safetensors` | WAI-REALISM | v1.0 | Fair AI Public License 1.0-SD | |
| |
| --- |
| |
| ## 📜 Credits & License |
| |
| ### 🏆 Special Acknowledgement |
| 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) |
| |
| ### Base Models |
| These models are derivatives of their respective creators. All credit for aesthetic tuning and model training belongs to the original creators. |
| - **JANKU Trained Chenkin & Noobai-Rouwei (Illustrious-XL)**: Created by [janxd](https://civitai.com/user/janxd). |
| - **blue_pencil-XL**: Created by [Euge_us](https://civitai.com/user/Euge_us). |
| - **epiCRealism XL**: Created by [epinikion](https://civitai.com/user/epinikion). |
| - **WAI-illustrious-SDXL / WAI-REAL_CN / WAI-REALISM**: Created by [WAI0731](https://civitai.com/user/WAI0731). |
| - **koronemixIllustrious / koronemixVpred**: Created by [koronen](https://civitai.com/user/koronen). |
| - **Nova Anime XL / Nova Asian XL**: Original creator on Civitai. |
| - **Prefect Illustrious XL**: Created by [Goofy_Ai](https://civitai.com/user/Goofy_Ai). |
| - **OneObsession**: Created by [Polyhedron](https://civitai.com/user/Polyhedron). |
| - **RealVisXL**: Created by [SG_161222](https://civitai.com/user/SG_161222). |
| - **UwazumiMix**: Created by [UWAZUMI](https://civitai.com/user/UWAZUMI). |
| |
| --- |
| |
| **Disclaimer:** These models are provided for optimization and research purposes. Please adhere to the original licenses of the base models. |
| |
