|
--- |
|
license: apache-2.0 |
|
tags: |
|
- T5xxl |
|
- Google FLAN |
|
--- |
|
|
|
# FLAN-T5-XXL Fused Model |
|
|
|
**Guide (External Site):** [English](https://www.ai-image-journey.com/2025/03/flan-t5xxl-te-only.html) | [Japanese](https://note.com/ai_image_journey/n/ncc6b1c475d8f) |
|
|
|
This repository hosts a fused version of the FLAN-T5-XXL model, created by combining the split files from [Google's FLAN-T5-XXL repository](https://huggingface.co/google/flan-t5-xxl). The files have been merged for convenience, making it easier to integrate into AI applications, including image generation workflows. |
|
|
|
<div style="display: flex; justify-content: center; align-items: center; gap: 2em;"> |
|
<div> |
|
<img src="./images/flan_t5_xxl_TE-only_FP32_sample1.png" alt="FLAN-T5-XXL sample image 1" width="400px" height="400px"> |
|
</div> |
|
<div> |
|
<img src="./images/flan_t5_xxl_TE-only_FP32_sample2.png" alt="FLAN-T5-XXL sample image 2" width="400px" height="400px"> |
|
</div> |
|
</div> |
|
|
|
Base Model: [**blue_pencil-flux1_v0.0.1**](https://huggingface.co/bluepen5805/blue_pencil-flux1) |
|
|
|
## Key Features |
|
|
|
- **Fused for Simplicity:** Combines split model files into a single, ready-to-use format. |
|
- **Optimized Variants:** Available in FP32, FP16, and quantized GGUF formats to balance accuracy and resource usage. |
|
- **Enhanced Prompt Accuracy:** Outperforms the standard T5-XXL v1.1 in generating precise outputs for image generation tasks. |
|
|
|
## Model Variants |
|
|
|
| Model | Size | SSIM Similality | Reccomend | |
|
| :---: | :---: | :---: | :---: | |
|
| FP32 | 19 GB | 100.0 % | 🔺 | |
|
| FP16 | 9.6 GB | 98.0 % | ✅ | |
|
| FP8 | 4.8 GB | 95.3 % | 🔺 | |
|
| Q8_0 | 6 GB | 97.6 % | ✅ | |
|
| Q6_K | 4.9 GB | 97.3 % | 🔺 | |
|
| Q5_K_M | 4.3 GB | 94.8 % | | |
|
| Q4_K_M | 3.7 GB | 96.4 % | | |
|
|
|
### Comparison Graph |
|
|
|
<div style="text-align: center; margin-left: auto; margin-right: auto; width: 600px; max-width: 80%;"> |
|
<img src="./images/Flan-T5xxl_TE-only_MAE_SSIM_Similarity.png" alt="Flan-T5xxl MAE and SSIM Similarity Graph"> |
|
</div> |
|
|
|
For a detailed comparison, refer to [this blog post](https://www.ai-image-journey.com/2024/12/image-difference-t5xxl-clip-l.html). |
|
|
|
## Usage Instructions |
|
|
|
Place the downloaded model files in one of the following directories: |
|
- `installation_folder/models/text_encoder` |
|
- `installation_folder/models/clip` |
|
- `installation_folder/Models/CLIP` |
|
|
|
### Stable Diffusion WebUI Forge |
|
|
|
In Stable Diffusion WebUI Forge, select the FLAN-T5-XXL model instead of the default T5xxl_v1_1 text encoder. |
|
|
|
<div style="text-align: center; margin-left: auto; margin-right: auto; width: 800px; max-width: 80%;"> |
|
<img src="./images/Screenshot of Stable Diffusion WebUI Forge text encoder selection screen.png" alt="Stable Diffusion WebUI Forge Text Encoder Selection Screen"> |
|
</div> |
|
|
|
**Note:** Stable Diffusion WebUI Forge does not support FP32 models. Use FP16 or GGUF formats instead. |
|
|
|
### ComfyUI |
|
|
|
**Sample Workflow** |
|
|
|
<div style="text-align: center; margin-left: auto; margin-right: auto; width: 1200px; max-width: 90%;"> |
|
<img src="./images/Flux1_MultiGPU 2025.3.20.png" alt="Flux1_MultiGPU sample workflow"> |
|
</div> |
|
|
|
<p style="text-align:center;font-size:0.8em;">This PNG image contains a ComfyUI workflow.</p> |
|
|
|
For ComfyUI, we recommend using the [ComfyUI-MultiGPU](https://github.com/neuratech-ai/ComfyUI-MultiGPU) custom node to load the model into system RAM instead of VRAM. |
|
|
|
<div style="text-align: center; margin-left: auto; margin-right: auto; width: 800px; max-width: 80%;"> |
|
<img src="./images/Screenshots of ComfyUI's DualCLIPLoaderMultiGPU and DualCLIPLoaderGGUFMultiGPU custom nodes.png" alt="ComfyUI DualCLIPLoaderMultiGPU and DualCLIPLoaderGGUFMultiGPU Custom Nodes"> |
|
</div> |
|
|
|
Use the **DualCLIPLoaderMultiGPU** or **DualCLIPLoaderGGUFMultiGPU** node and set the device to **cpu** to load the model into system RAM. |
|
|
|
**FP32 Support:** To use FP32 text encoders in ComfyUI, launch with the `--fp32-text-enc` flag. |
|
|
|
## Comparison: FLAN-T5-XXL vs T5-XXL v1.1 |
|
|
|
<div style="display: flex; justify-content: center; align-items: center; gap: 2em;"> |
|
<div> |
|
<img src="./images/flan_t5_xxl_image.png" alt="FLAN-T5-XXL Image" width="400px" height="400px"> |
|
</div> |
|
<div> |
|
<img src="./images/t5_xxl_v1_1_image.png" alt="T5-XXL v1.1 Image" width="400px" height="400px"> |
|
</div> |
|
</div> |
|
|
|
These example images were generated using **FLAN-T5-XXL** and [**T5-XXL v1.1**](https://huggingface.co/google/t5-v1_1-xxl) models in Flux.1. FLAN-T5-XXL delivers more accurate responses to prompts. |
|
|
|
## Further Comparisons |
|
|
|
- [FLAN-T5-XXL vs T5-XXL v1.1](https://www.ai-image-journey.com/2024/12/clip-t5xxl-text-encoder.html) |
|
- [FLAN-T5-XXL FP32 vs FP16 and Quantization](https://www.ai-image-journey.com/2024/12/image-difference-t5xxl-clip-l.html) |
|
|
|
--- |
|
|
|
## License |
|
|
|
- This model is distributed under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0). |
|
- The uploader claims no ownership or rights over the model. |
