Hugging Face's logo

THUDM
/
CogView4-6B

Text-to-Image
Diffusers
Safetensors
Chinese
English
CogView4Pipeline
Model card Files Community
6
Yuxuan Zhang commited on
Commit
71306a5
·
1 Parent(s): fb6f572

update

Browse files
Files changed (1) hide show
  1. README.md +53 -51
README.md CHANGED
@@ -1,10 +1,10 @@
1
  ---
2
  license: apache-2.0
3
  language:
4
- - zh
5
- - en
6
  base_model:
7
- - THUDM/glm-4-9b
8
  pipeline_tag: text-to-image
9
  library_name: diffusers
10
  ---
@@ -25,18 +25,19 @@ library_name: diffusers
25
 
26
  ## Inference Requirements and Model Introduction
27
 
28
- + Resolution: Width and height must be between `512px` and `2048px`, divisible by `32`, and ensure the maximum number of pixels does not exceed `2^21` px.
 
29
  + Precision: BF16 / FP32 (FP16 is not supported as it will cause overflow resulting in completely black images)
30
 
31
  Using `BF16` precision with `batchsize=4` for testing, the memory usage is shown in the table below:
32
 
33
- | Resolution | enable_model_cpu_offload OFF | enable_model_cpu_offload ON | enable_model_cpu_offload ON </br> Text Encoder 4bit |
34
- |--------------|------------------------------|-----------------------------|----------------------------------------------------|
35
- | 512 * 512 | 33GB | 20GB | 13G |
36
- | 1280 * 720 | 35GB | 20GB | 13G |
37
- | 1024 * 1024 | 35GB | 20GB | 13G |
38
- | 1920 * 1280 | 39GB | 20GB | 14G |
39
- | 2048 * 2048 | 43GB | 21GB | 14G |
40
 
41
  ## Quick Start
42
 
@@ -52,6 +53,7 @@ Then, run the following code:
52
 
53
  ```python
54
  from diffusers import CogView4Pipeline
 
55
  pipe = CogView4Pipeline.from_pretrained("THUDM/CogView4-6B", torch_dtype=torch.bfloat16)
56
 
57
  # Open it for reduce GPU memory usage
@@ -72,52 +74,52 @@ image = pipe(
72
  image.save("cogview4.png")
73
  ```
74
 
75
- ## Model Performance
76
 
77
  We've tested on multiple benchmarks and achieved the following scores:
78
 
79
- ### dpg_bench
80
-
81
- | model | overall | global | entity | attribute | relation | other |
82
- |-------|---------|--------|--------|-----------|----------|-------|
83
- | sdxl | 74.65 | 83.27 | 82.43 | 80.91 | 86.76 | 80.41 |
84
- | pixart-alpha | 71.11 | 74.97 | 79.32 | 78.60 | 82.57 | 76.96 |
85
- | sd3-medium | 84.08 | 87.90 | **91.01** | 88.83 | 80.70 | 88.68 |
86
- | dalle-3 | 83.50 | **90.97** | 89.61 | 88.39 | 90.58 | 89.83 |
87
- | flux.1-dev | 83.79 | 85.80 | 86.79 | 89.98 | 90.04 | **89.90** |
88
- | **cogview4** | **85.13** | 83.85 | 90.35 | **91.17** | **91.14** | 87.29 |
89
-
90
-
91
- ### Geneval
92
-
93
- | model | overall | single | two | counting | colors | position | Color attribution |
94
- |-------|---------|--------|-----|----------|--------|----------|------------------|
95
- | sdxl | 0.55 | 0.98 | 0.74 | 0.39 | 0.85 | 0.15 | 0.23 |
96
- | pixart-alpha | 0.48 | 0.98 | 0.50 | 0.44 | 0.80 | 0.08 | 0.07 |
97
- | sd3-meidum | **0.74** | **0.99** | **0.94** | **0.72** | **0.89** | 0.33 | **0.60** |
98
- | dall-e 3 | 0.67 | 0.96 | 0.87 | 0.47 | 0.83 | 0.43 | 0.45 |
99
- | flux.1-dev | 0.66 | 0.98 | 0.79 | 0.73 | 0.77 | 0.22 | 0.45 |
100
- | **cogview4** | 0.73 | **0.99** | 0.86 | 0.66 | 0.79 | **0.48** | 0.58 |
101
-
102
- ### t2i_compbench
103
-
104
- | model | color | shape | texture | 2d-spatial | 3d-spatial | numeracy | Non-spatial clip | complex 3-in-1 |
105
- |-------|-------|-------|---------|------------|------------|----------|-----------------|---------------|
106
- | sdxl | 0.5879 | 0.4687 | 0.5299 | 0.2133 | 0.3566 | 0.4988 | 0.3119 | 0.3237 |
107
- | pixart-alpha | 0.6690 | 0.4927 | 0.6477 | 0.2064 | 0.3901 | 0.5058 | **0.3197** | 0.3433 |
108
- | sd3-medium | **0.8132** | 0.5885 | **0.7334** | **0.3200** | **0.4084** | 0.6174 | 0.3140 | 0.3771 |
109
- | dall-e 3 | 0.7785 | **0.6205** | 0.7036 | 0.2865 | 0.3744 | 0.5880 | 0.3003 | **0.3773** |
110
- | flux.1-dev | 0.7572 | 0.5066 | 0.6300 | 0.2700 | 0.3992 | 0.6165 | 0.3065 | 0.3628 |
111
- | **cogview4** | 0.7786 | 0.5880 | 0.6983 | 0.3075 | 0.3708 | **0.6626** | 0.3056 | 0.3869 |
112
-
 
113
 
114
  ## Chinese Text Accuracy Evaluation
115
 
116
- | model | Precision | Recall | F1 Score | pick@4 |
117
- |-------|-----------|--------|----------|--------|
118
- | kolors | 0.6094 | 0.1886 | 0.2880 | 0.1633 |
119
- | **cogview4** | **0.6969** | **0.5532** | **0.6168** | **0.3265** |
120
-
121
 
122
  ## Citation
123
 
 
1
  ---
2
  license: apache-2.0
3
  language:
4
+ - zh
5
+ - en
6
  base_model:
7
+ - THUDM/glm-4-9b
8
  pipeline_tag: text-to-image
9
  library_name: diffusers
10
  ---
 
25
 
26
  ## Inference Requirements and Model Introduction
27
 
28
+ + Resolution: Width and height must be between `512px` and `2048px`, divisible by `32`, and ensure the maximum number of
29
+ pixels does not exceed `2^21` px.
30
  + Precision: BF16 / FP32 (FP16 is not supported as it will cause overflow resulting in completely black images)
31
 
32
  Using `BF16` precision with `batchsize=4` for testing, the memory usage is shown in the table below:
33
 
34
+ | Resolution | enable_model_cpu_offload OFF | enable_model_cpu_offload ON | enable_model_cpu_offload ON </br> Text Encoder 4bit |
35
+ |-------------|------------------------------|-----------------------------|-----------------------------------------------------|
36
+ | 512 * 512 | 33GB | 20GB | 13G |
37
+ | 1280 * 720 | 35GB | 20GB | 13G |
38
+ | 1024 * 1024 | 35GB | 20GB | 13G |
39
+ | 1920 * 1280 | 39GB | 20GB | 14G |
40
+ | 2048 * 2048 | 43GB | 21GB | 14G |
41
 
42
  ## Quick Start
43
 
 
53
 
54
  ```python
55
  from diffusers import CogView4Pipeline
56
+
57
  pipe = CogView4Pipeline.from_pretrained("THUDM/CogView4-6B", torch_dtype=torch.bfloat16)
58
 
59
  # Open it for reduce GPU memory usage
 
74
  image.save("cogview4.png")
75
  ```
76
 
77
+ ### Model Metrics
78
 
79
  We've tested on multiple benchmarks and achieved the following scores:
80
 
81
+ #### dpg_bench
82
+
83
+ | Model | Overall | Global | Entity | Attribute | Relation | Other |
84
+ |--------------|-----------|-----------|-----------|-----------|-----------|-----------|
85
+ | SDXL | 74.65 | 83.27 | 82.43 | 80.91 | 86.76 | 80.41 |
86
+ | PixArt-alpha | 71.11 | 74.97 | 79.32 | 78.60 | 82.57 | 76.96 |
87
+ | SD3-Medium | 84.08 | 87.90 | **91.01** | 88.83 | 80.70 | 88.68 |
88
+ | DALL-E 3 | 83.50 | **90.97** | 89.61 | 88.39 | 90.58 | 89.83 |
89
+ | Flux.1-dev | 83.79 | 85.80 | 86.79 | 89.98 | 90.04 | **89.90** |
90
+ | Janus-Pro-7B | 84.19 | 86.90 | 88.90 | 89.40 | 89.32 | 89.48 |
91
+ | **cogview4** | **85.13** | 83.85 | 90.35 | **91.17** | **91.14** | 87.29 |
92
+
93
+ #### Geneval
94
+
95
+ | Model | Overall | Single Obj. | Two Obj. | Counting | Colors | Position | Color attribution |
96
+ |-----------------|----------|-------------|----------|----------|----------|----------|-------------------|
97
+ | SDXL | 0.55 | 0.98 | 0.74 | 0.39 | 0.85 | 0.15 | 0.23 |
98
+ | PixArt-alpha | 0.48 | 0.98 | 0.50 | 0.44 | 0.80 | 0.08 | 0.07 |
99
+ | SD3-Medium | 0.74 | **0.99** | **0.94** | 0.72 | 0.89 | 0.33 | 0.60 |
100
+ | DALL-E 3 | 0.67 | 0.96 | 0.87 | 0.47 | 0.83 | 0.43 | 0.45 |
101
+ | Flux.1-dev | 0.66 | 0.98 | 0.79 | **0.73** | 0.77 | 0.22 | 0.45 |
102
+ | Janus-Pro-7B | **0.80** | **0.99** | 0.89 | 0.59 | **0.90** | **0.79** | **0.66** |
103
+ | **CogView4-6B** | 0.73 | **0.99** | 0.86 | 0.66 | 0.79 | 0.48 | 0.58 |
104
+
105
+ #### t2i_compbench
106
+
107
+ | Model | Color | Shape | Texture | 2D-Spatial | 3D-Spatial | Numeracy | Non-spatial Clip | Complex 3-in-1 |
108
+ |-----------------|------------|------------|------------|------------|------------|------------|------------------|----------------|
109
+ | SDXL | 0.5879 | 0.4687 | 0.5299 | 0.2133 | 0.3566 | 0.4988 | 0.3119 | 0.3237 |
110
+ | PixArt-alpha | 0.6690 | 0.4927 | 0.6477 | 0.2064 | 0.3901 | 0.5058 | **0.3197** | 0.3433 |
111
+ | SD3-Medium | **0.8132** | 0.5885 | **0.7334** | **0.3200** | **0.4084** | 0.6174 | 0.3140 | 0.3771 |
112
+ | DALL-E 3 | 0.7785 | **0.6205** | 0.7036 | 0.2865 | 0.3744 | 0.5880 | 0.3003 | 0.3773 |
113
+ | Flux.1-dev | 0.7572 | 0.5066 | 0.6300 | 0.2700 | 0.3992 | 0.6165 | 0.3065 | 0.3628 |
114
+ | Janus-Pro-7B | 0.5145 | 0.3323 | 0.4069 | 0.1566 | 0.2753 | 0.4406 | 0.3137 | 0.3806 |
115
+ | **CogView4-6B** | 0.7786 | 0.5880 | 0.6983 | 0.3075 | 0.3708 | **0.6626** | 0.3056 | **0.3869** |
116
 
117
  ## Chinese Text Accuracy Evaluation
118
 
119
+ | model | Precision | Recall | F1 Score | pick@4 |
120
+ |-----------------|------------|------------|------------|------------|
121
+ | kolors | 0.6094 | 0.1886 | 0.2880 | 0.1633 |
122
+ | **CogView4-6B** | **0.6969** | **0.5532** | **0.6168** | **0.3265** |
 
123
 
124
  ## Citation
125