File size: 4,777 Bytes
7865f12 6dae96c e06259e 7865f12 6dae96c 7865f12 6dae96c 7865f12 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 | ---
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 |