Hugging Face's logo

easygoing0114
/
flan-t5-xxl-fused

GGUF
T5xxl
Google FLAN
Inference Endpoints
Model card Files Community
1
easygoing0114 commited on
Commit
fdbfc65
·
verified ·
1 Parent(s): e08dad9

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +75 -46
README.md CHANGED
@@ -1,84 +1,113 @@
1
  ---
2
  license: apache-2.0
3
  tags:
4
- - T5xxl
5
- - Google FLAN
6
  ---
7
 
8
  # FLAN-T5-XXL Fused Model
9
 
10
- This repository contains a fused version of FLAN-T5-XXL, combining the split files available at [Google's FLAN-T5-XXL repository](https://huggingface.co/google/flan-t5-xxl). These files have been merged for easier use in AI applications, including image generation models.
11
 
12
- ---
13
-
14
- ## Newly Added: TE-Only Models for Stable Diffusion WebUI Forge & ComfyUI (2025-03-04)
15
- Two additional files provide the **Text Encoder (TE) only** portion of FLAN-T5-XXL, specifically extracted for use with Stable Diffusion WebUI Forge and ComfyUI.
16
-
17
- <div style="display: flex; justify-content: center; align-items: center;">
18
- <div style="text-align: center; margin-right: 1em;">
19
- <img src="./flan_t5_xxl_TE-only_FP32_sample1.png" alt="flan_t5_xxl_TE-only_FP32_sample1" width="400px" height="400px">
20
  </div>
21
- <div style="text-align: center; margin-left: 1em;">
22
- <img src="./flan_t5_xxl_TE-only_FP32_sample2.png" alt="flan_t5_xxl_TE-only_FP32_sample2" width="400px" height="400px">
23
  </div>
24
  </div>
25
 
26
- - `flan_t5_xxl_TE-only_FP32.safetensors`: Full-precision FP32 TE-only model.
27
- - `flan_t5_xxl_TE-only_FP16.safetensors`: Half-precision FP16 TE-only model for memory-efficient inference.
28
 
29
- These models retain only the text encoding functionality of FLAN-T5-XXL, reducing resource consumption while maintaining high-quality prompt processing in AI image generation workflows.
30
 
31
- Also can be used as drop-in replacements for standard text encoders in Stable Diffusion-based workflows.
 
 
32
 
 
33
 
34
- ---
 
 
 
 
 
 
 
 
 
 
35
 
36
- ## Full Models
37
- - `flan_t5_xxl_fp32.safetensors`: Full-precision FP32 Full model.
38
- - `flan_t5_xxl_fp16.safetensors`: Half-precision FP16 Full model for memory-efficient inference.
39
 
40
- ---
 
 
41
 
42
- ## Comparison: FLAN-T5-XXL-FP32 vs FLAN-T5-XXL-FP16 on Flux.1[dev] (base model: [blue_pencil-flux1_v0.0.1](https://huggingface.co/bluepen5805/blue_pencil-flux1))
43
 
44
- <div style="text-align: center; margin-left: auto; margin-right: auto;width:600px;max-width:80%;">
45
- <img src="./Flan-T5xxl-FP32_FP16_compare.png" alt="Flan-T5xxl-FP32_FP16_compare">
 
 
 
 
 
 
 
 
 
 
 
46
  </div>
47
 
48
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
49
 
50
  ## Comparison: FLAN-T5-XXL vs T5-XXL v1.1
51
 
52
- <div style="display: flex; justify-content: center; align-items: center;">
53
- <div style="text-align: center; margin-right: 1em;">
54
- <img src="./flan_t5_xxl_image.png" alt="FLAN-T5-XXL Image" width="400px" height="400px">
55
- <p>FLAN-T5-XXL Output</p>
56
  </div>
57
- <div style="text-align: center; margin-left: 1em;">
58
- <img src="./t5_xxl_v1_1_image.png" alt="T5-XXL v1.1 Image" width="400px" height="400px">
59
- <p>T5-XXL v1.1 Output</p>
60
  </div>
61
  </div>
62
 
63
- These example images generated using **FLAN-T5-XXL** and [**T5-XXL v1.1**](https://huggingface.co/google/t5-v1_1-xxl) models in Flux.1.
64
 
65
- FLAN-T5-XXL provides more accurate responses to prompts.
66
 
67
- ---
 
 
 
68
 
69
- ## Further Comparison
 
 
70
 
71
- - [FLAN-T5-XXL vs T5-XXL v1.1](https://www.ai-image-journey.com/2024/12/clip-t5xxl-text-encoder.html)
72
- - [FLAN-T5-XXL FP32 vs FP16 and other quantization](https://www.ai-image-journey.com/2024/12/image-difference-t5xxl-clip-l.html)
73
 
74
  ---
75
 
76
  ## License
77
- This model is provided under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0).
78
- The uploader does not claim any rights over the model.
79
-
80
- ---
81
 
82
- ## Acknowledgments
83
- - Original source: [Google's FLAN-T5-XXL repository](https://huggingface.co/google/flan-t5-xxl).
84
- - GGUF version: [dumb-dev's Hugging Face repository](https://huggingface.co/dumb-dev/flan-t5-xxl-gguf).
 
1
  ---
2
  license: apache-2.0
3
  tags:
4
+ - T5xxl
5
+ - Google FLAN
6
  ---
7
 
8
  # FLAN-T5-XXL Fused Model
9
 
10
+ 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.
11
 
12
+ <div style="display: flex; justify-content: center; align-items: center; gap: 2em;">
13
+ <div>
14
+ <img src="./images/flan_t5_xxl_TE-only_FP32_sample1.png" alt="FLAN-T5-XXL sample image 1" width="400px" height="400px">
 
 
 
 
 
15
  </div>
16
+ <div>
17
+ <img src="./images/flan_t5_xxl_TE-only_FP32_sample2.png" alt="FLAN-T5-XXL sample image 2" width="400px" height="400px">
18
  </div>
19
  </div>
20
 
21
+ Sample pictures: Base Model [**blue_pencil-flux1_v0.0.1**](https://huggingface.co/bluepen5805/blue_pencil-flux1)
 
22
 
23
+ ## Key Features
24
 
25
+ - **Fused for Simplicity:** Combines split model files into a single, ready-to-use format.
26
+ - **Optimized Variants:** Available in FP32, FP16, and quantized GGUF formats to balance accuracy and resource usage.
27
+ - **Enhanced Prompt Accuracy:** Outperforms the standard T5-XXL v1.1 in generating precise outputs for image generation tasks.
28
 
29
+ ## Model Variants
30
 
31
+ | Model File | Size | Accuracy (SSIM Similarity) | Recommended |
32
+ |-------------------------------------|:--------:|:----------------------------:|:-------------:|
33
+ | flan_t5_xxl_fp32.safetensors | 44.1GB | 100% | |
34
+ | flan_t5_xxl_fp16.safetensors | 22.1GB | 99.9% | |
35
+ | flan_t5_xxl_TE-only_FP32.safetensors| 18.7GB | 100% | 🔺 |
36
+ | flan_t5_xxl_TE-only_FP16.safetensors| 9.4GB | 99.9% | ✅ |
37
+ | flan_t5_xxl_TE-only_Q8_0.gguf | 5.5GB | 99.8% | ✅ |
38
+ | flan_t5_xxl_TE-only_Q6_K.gguf | 4.4GB | 99.7% | 🔺 |
39
+ | flan_t5_xxl_TE-only_Q5_K_M.gguf | 3.8GB | 98.4% | 🔺 |
40
+ | flan_t5_xxl_TE-only_Q4_K_M.gguf | 3.2GB | 95.2% | |
41
+ | flan_t5_xxl_TE-only_Q3_K_L.gguf | 2.6GB | 84.9% | |
42
 
43
+ ### Comparison Graph
 
 
44
 
45
+ <div style="text-align: center; margin-left: auto; margin-right: auto; width: 600px; max-width: 80%;">
46
+ <img src="./images/Flan-T5xxl MAE and SSIM Similarity.png" alt="Flan-T5xxl MAE and SSIM Similarity Graph">
47
+ </div>
48
 
49
+ For a detailed comparison, refer to [this blog post](https://www.ai-image-journey.com/2024/12/image-difference-t5xxl-clip-l.html).
50
 
51
+ ## Usage Instructions
52
+
53
+ Place the downloaded model files in one of the following directories:
54
+ - `installation_folder/models/text_encoder`
55
+ - `installation_folder/models/clip`
56
+ - `installation_folder/Models/CLIP`
57
+
58
+ ### Stable Diffusion WebUI Forge
59
+
60
+ In Stable Diffusion WebUI Forge, select the FLAN-T5-XXL model instead of the default T5xxl_v1_1 text encoder.
61
+
62
+ <div style="text-align: center; margin-left: auto; margin-right: auto; width: 800px; max-width: 80%;">
63
+ <img src="./images/Screenshot of Stable Diffusion WebUI Forge text encoder selection screen.png" alt="Stable Diffusion WebUI Forge Text Encoder Selection Screen">
64
  </div>
65
 
66
+ **Note:** Stable Diffusion WebUI Forge does not support FP32 models. Use FP16 or GGUF formats instead.
67
+
68
+ ### ComfyUI
69
+
70
+ **Sample Workflow**
71
+
72
+ 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.
73
+
74
+ <div style="text-align: center; margin-left: auto; margin-right: auto; width: 800px; max-width: 80%;">
75
+ <img src="./images/Screenshots of ComfyUI's DualCLIPLoaderMultiGPU and DualCLIPLoaderGGUFMultiGPU custom nodes.png" alt="ComfyUI DualCLIPLoaderMultiGPU and DualCLIPLoaderGGUFMultiGPU Custom Nodes">
76
+ </div>
77
+
78
+ Use the **DualCLIPLoaderMultiGPU** or **DualCLIPLoaderGGUFMultiGPU** node and set the device to **cpu** to load the model into system RAM.
79
+
80
+ **FP32 Support:** To use FP32 text encoders in ComfyUI, launch with the `--fp32-text-enc` flag.
81
 
82
  ## Comparison: FLAN-T5-XXL vs T5-XXL v1.1
83
 
84
+ <div style="display: flex; justify-content: center; align-items: center; gap: 2em;">
85
+ <div>
86
+ <img src="./images/flan_t5_xxl_image.png" alt="FLAN-T5-XXL Image" width="400px" height="400px">
 
87
  </div>
88
+ <div>
89
+ <img src="./images/t5_xxl_v1_1_image.png" alt="T5-XXL v1.1 Image" width="400px" height="400px">
 
90
  </div>
91
  </div>
92
 
93
+ 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.
94
 
95
+ ## Further Comparisons
96
 
97
+ - [FLAN-T5-XXL vs T5-XXL v1.1](https://ai-image-journey.blogspot.com/2024/12/clip-t5xxl-text-encoder.html)
98
+ - [FLAN-T5-XXL FP32 vs FP16 and Quantization](https://ai-image-journey.blogspot.com/2024/12/image-difference-t5xxl-clip-l.html)
99
+
100
+ ### Tip: Upgrade CLIP-L Too
101
 
102
+ For even better results, consider upgrading the CLIP-L text encoder alongside FLAN-T5-XXL:
103
+ - [LongCLIP-SAE-ViT-L-14](https://huggingface.co/zer0int/LongCLIP-SAE-ViT-L-14) (ComfyUI only)
104
+ - [CLIP-SAE-ViT-L-14](https://huggingface.co/zer0int/CLIP-SAE-ViT-L-14)
105
 
106
+ Combining FLAN-T5-XXL with an upgraded CLIP-L can further enhance image quality.
 
107
 
108
  ---
109
 
110
  ## License
 
 
 
 
111
 
112
+ - This model is distributed under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0).
113
+ - The uploader claims no ownership or rights over the model.