Hugging Face's logo

Bedovyy
/
ERNIE-Image-Quantized

Text-to-Image
Diffusion Single File
comfyui
Model card Files
xet
 Community
4
Bedovyy commited on
Commit
925ae0f
·
verified ·
1 Parent(s): c757b3e

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +48 -7
README.md CHANGED
@@ -15,16 +15,45 @@ pipeline_tag: text-to-image
15
  - fp8e4m3 (8.22GB)
16
  - int8rowwise (8.22GB) - it needs [ComfyUI-INT8-Fast](https://github.com/BobJohnson24/ComfyUI-INT8-Fast) custom node
17
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
18
 
19
- ## Sample
20
 
21
- ### ERNIE-Image
22
 
23
- ![ernie-image-comp1](https://cdn-uploads.huggingface.co/production/uploads/63fbf6951b4b1bd4e706fed1/Hy24cvgMbDFsvTgQeXw5Z.webp)
24
-
25
- ![ernie-image-comp2](https://cdn-uploads.huggingface.co/production/uploads/63fbf6951b4b1bd4e706fed1/Ro03949AktXAEqzJcmder.webp)
26
-
27
- ![ernie-image-comp3](https://cdn-uploads.huggingface.co/production/uploads/63fbf6951b4b1bd4e706fed1/hN0Ow5MRznmTONz20Nyr5.webp)
28
 
29
  ### ERNIE-Image-Turbo
30
 
@@ -34,6 +63,18 @@ pipeline_tag: text-to-image
34
 
35
  ![ernie-image-turbo-comp3](https://cdn-uploads.huggingface.co/production/uploads/63fbf6951b4b1bd4e706fed1/QARRTSNDpe-4f4DAjX-_2.webp)
36
 
 
 
 
 
 
 
 
 
 
 
 
 
37
 
38
  ## How to reproduce
39
 
 
15
  - fp8e4m3 (8.22GB)
16
  - int8rowwise (8.22GB) - it needs [ComfyUI-INT8-Fast](https://github.com/BobJohnson24/ComfyUI-INT8-Fast) custom node
17
 
18
+ ## Generation Speed
19
+
20
+ ### ERINE-Image-Turbo
21
+
22
+ | GPU | Quantization | Speed (it/s) | Time (secs) | vs BF16 |
23
+ |----------|--------------|--------------|-------------|---------------------|
24
+ | RTX 5090 | bf16 | 2.09 | 4.87 | 100% |
25
+ | | fp8e4m3 | 3.69 | 3.32 | 147% |
26
+ | | int8rowwise | 4.31 | 3.05 | *160%* |
27
+ | | nvfp4 | 5.09 | 2.72 | **179%** |
28
+ | RTX 3090 | bf16 | 0.88 | 12.42 | *100%* |
29
+ | | fp8e4m3 | 0.84 | 12.73 | 98% |
30
+ | | int8rowwise | 1.66 | 7.04 | **176%** |
31
+ | | nvfp4 | 0.83 | 12.71 | 98% |
32
+ | RTX 3060 | bf16 | 0.26 | 43.02 | 100% |
33
+ | | fp8e4m3 | 0.39 | 28.66 | *150%* |
34
+ | | int8rowwise | 0.82 | 14.43 | **298%** |
35
+ | | nvfp4 | 0.39 | 28.72 | 150% |
36
+
37
+ ### ERINE-Image
38
+
39
+ | GPU | Quantization | Speed (it/s) | Time (secs) | vs BF16 |
40
+ |----------|--------------|--------------|-------------|---------------------|
41
+ | RTX 5090 | bf16 | 1.08 | 20.08 | 100% |
42
+ | | fp8e4m3 | 1.97 | 11.67 | 172% |
43
+ | | int8rowwise | 2.14 | 10.89 | *184%* |
44
+ | | nvfp4 | 2.56 | 9.35 | **215%** |
45
+ | RTX 3090 | bf16 | 0.40 | 53.33 | *100%* |
46
+ | | fp8e4m3 | 0.39 | 54.71 | 97% |
47
+ | | int8rowwise | 0.79 | 28.08 | **190%** |
48
+ | | nvfp4 | 0.38 | 55.20 | 97% |
49
+ | RTX 3060 | bf16 | 0.11 | 201.41 | 100% |
50
+ | | fp8e4m3 | 0.17 | 130.48 | *154%* |
51
+ | | int8rowwise | 0.35 | 62.42 | **323%** |
52
+ | | nvfp4 | 0.17 | 130.87 | 154% |
53
 
 
54
 
 
55
 
56
+ ## Sample
 
 
 
 
57
 
58
  ### ERNIE-Image-Turbo
59
 
 
63
 
64
  ![ernie-image-turbo-comp3](https://cdn-uploads.huggingface.co/production/uploads/63fbf6951b4b1bd4e706fed1/QARRTSNDpe-4f4DAjX-_2.webp)
65
 
66
+ ![erine-image-turbo-comp4](https://cdn-uploads.huggingface.co/production/uploads/63fbf6951b4b1bd4e706fed1/lz3daUybkP-ggSNtV0Nkv.webp)
67
+
68
+ ### ERNIE-Image
69
+
70
+ ![ernie-image-comp1](https://cdn-uploads.huggingface.co/production/uploads/63fbf6951b4b1bd4e706fed1/Hy24cvgMbDFsvTgQeXw5Z.webp)
71
+
72
+ ![ernie-image-comp2](https://cdn-uploads.huggingface.co/production/uploads/63fbf6951b4b1bd4e706fed1/Ro03949AktXAEqzJcmder.webp)
73
+
74
+ ![ernie-image-comp3](https://cdn-uploads.huggingface.co/production/uploads/63fbf6951b4b1bd4e706fed1/hN0Ow5MRznmTONz20Nyr5.webp)
75
+
76
+ ![erine-image-comp4](https://cdn-uploads.huggingface.co/production/uploads/63fbf6951b4b1bd4e706fed1/6ZZ3_Cj1brN3cQ4dDPpGv.webp)
77
+
78
 
79
  ## How to reproduce
80