| --- |
| license: other |
| tags: |
| - quantized |
| - int8 |
| - comfyui |
| - diffusion_models |
| - convrot |
| - text_encoders |
| --- |
| |
| # 📦 INT8 Quantized Model Collection (ConvRot) |
|
|
| This repository contains a collection of popular generative models quantized to **INT8** using the **ConvRot** method for optimized inference with minimal quality loss. |
|
|
| Almost all models were quantized from the original **BF16** weights using the `quant_int8_convrot.py` script with the `--mseclip` attribute for calibration. |
|
|
| --- |
|
|
| ## 🛠️ Quantization Details |
|
|
| - **Base Precision:** BF16 → INT8 ConvRot |
| - **Method:** ConvRot (Convolution Rotation) |
| - **Calibration:** `--mseclip` for per-layer clipping optimization (couple has `--absmax`) |
| - **Tool:** [`quant_int8_convrot.py`](https://github.com/Comfy-Org/comfy-model-tools/blob/main/quant_int8_convrot.py) (Comfy-Org) |
|
|
| `flux-2-dev_int8_convrot_quantops.safetensors` was first quantized in ComfyUI using [ComfyUI-QuantOps](https://github.com/silveroxides/ComfyUI-QuantOps), then convert to `int8_convrot` format via [`convert_to_comfy.py`](https://github.com/BobJohnson24/ComfyUI-INT8-Fast/blob/main/convert_to_comfy.py) |
|
|
| --- |
|
|
| ## 📂 Included Models |
|
|
| | Model Name | Original Source | Notes | |
| |------------|----------------|-------| |
| |**Checkpoints** |
| |`ltx-2.3-22b-dev_int8_convrot.safetensors` | LTX 2.3 22B | — | |
| | |
| |**Diffusion models** |
| | `flux-2-dev_int8_convrot_quantops.safetensors` | Flux-2 dev | via ComfyUI-INT8-Fast | |
| | `flux-2-dev_int8_convrot_mseclip.safetensors` | Flux-2 dev | — | |
| | `flux-2-dev_int8_convrot_absmax.safetensors` | Flux-2 dev | — | |
| | `flux-2-klein-4b_int8_convrot.safetensors` | Flux-2 Klein 4B | — | |
| | `flux-2-klein-9b_int8_convrot.safetensors` | Flux-2 Klein 9B | — | |
| | `flux-2-klein-9b-kv_int8_convrot.safetensors` | Flux-2 Klein 9B KV | — | |
| | `qwen_image_2512_int8_convrot.safetensors` | Qwen Image 2512 | — | |
| | `qwen_image_edit_2511_int8_convrot.safetensors` | Qwen Image Edit 2511 | — | |
| | `z_image_int8_convrot.safetensors` | Z-Image | — | |
| | `z_image_turbo_int8_convrot.safetensors` | Z-Image Turbo | — | |
| | `krea2_raw_int8_convrot.safetensors` | Krea 2 Raw | — | |
| | `krea2_turbo_int8_convrot.safetensors` | Krea 2 Turbo | — | |
| | `ideogram4_int8_convrot.safetensors` | Ideogram 4 | via dequant FP8 | |
| | `ideogram4_unconditional_int8_convrot.safetensors` | Ideogram 4 Uncond. | via dequant FP8 | |
| | |
| |**3D models** |
| | `trellis_2_int8_convrot.safetensors` | Trellis 2 | — | |
| | `pixal3d_int8_convrot.safetensors` | Pixal 3D | — | |
| | |
| |**Text encoders** |
| | `gemma4_e4b_it_int8_convrot.safetensors` | Gemma 4 E4B IT | — | |
| | `gemma_2_2b_it_elm_int8_convrot.safetensors` | Gemma 2 2B IT | — | |
| | `qwen3.5_4b_int8_convrot.safetensors` | Qwen 3.5 4B | — | |
| | `qwen3vl_8b_int8_convrot.safetensors` | Qwen 3 VL 8B | — | |
| | `qwen3vl_4b_int8_convrot.safetensors` | Qwen 3 VL 4B | — | |
| | `qwen_3_8b_int8_convrot.safetensors` | Qwen 3 8B | — | |
| | `qwen_3_4b_int8_convrot.safetensors` | Qwen 3 4B | — | |
| | `qwen_2.5_vl_7b_int8_convrot.safetensors` | Qwen 2.5 VL 7B | — | |
| | `mistral_3_small_flux2_int8_convrot.safetensors` | Mistral 3 Small for Flux2 | — | |
|
|
|
|
| > **Note:** The Ideogram 4 models were sourced from [Comfy-Org/Ideogram-4](https://huggingface.co/Comfy-Org/Ideogram-4) and added to this repository for convenience purposes. |
|
|
| --- |
|
|
| ## 📜 Licenses |
|
|
| Each model retains the **original license** from its respective base model. Please refer to the original model cards for licensing terms: |
|
|
| - [LTX 2.3](https://huggingface.co/Lightricks/LTX-2.3) |
| - |
| - [Flux.2 dev](https://huggingface.co/black-forest-labs/FLUX.2-dev) |
| - [Flux.2 klein 9b](https://huggingface.co/black-forest-labs/FLUX.2-klein-9B) |
| - [Flux.2 klein 9b-kv](https://huggingface.co/black-forest-labs/FLUX.2-klein-9b-kv) |
| - [Flux.2 klein 4b](https://huggingface.co/black-forest-labs/FLUX.2-klein-4B) |
| - [Qwen-Image-2512](https://huggingface.co/Qwen/Qwen-Image-2512) |
| - [Qwen-Image-Edit-2511](https://huggingface.co/Qwen/Qwen-Image-Edit-2511) |
| - [Z-Image](https://huggingface.co/Tongyi-MAI/Z-Image) |
| - [Z-Image Turbo](https://huggingface.co/Tongyi-MAI/Z-Image-Turbo) |
| - [Krea 2](https://huggingface.co/krea/Krea-2-Raw) |
| - [Ideogram 4](https://huggingface.co/ideogram-ai/ideogram-4-fp8) |
| - |
| - [Trellis.2](https://huggingface.co/microsoft/TRELLIS.2-4B) |
| - [Pixal3D](https://huggingface.co/TencentARC/Pixal3D) |
| - |
| - [Gemma](https://huggingface.co/google/) |
| - [Qwen](https://huggingface.co/Qwen) |
| - [Mistral](https://huggingface.co/mistralai) |
|
|
| --- |
|
|
| ## 🙏 Acknowledgments |
|
|
| - [Comfy-Org](https://github.com/Comfy-Org) for the quantization script and bf16 models |
| - [BobJohnson24](https://github.com/BobJohnson24) for `ComfyUI-INT8-Fast` and conversion tools |
| - [silveroxides](https://github.com/silveroxides) for `ComfyUI-QuantOps` nodes |
| - All original model authors and teams |
