MonsterMMORPG commited on
Commit
1db2427
1 Parent(s): 0c3c3c4

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +30 -4
README.md CHANGED
@@ -1,4 +1,4 @@
1
- This is a training of a public LoRA style (2 seperate training each on 4x A6000).
2
 
3
  Experimenting captions vs non-captions. So we will see which yields best results for style training on FLUX.
4
 
@@ -23,11 +23,37 @@ The no caption dataset uses only ohwx 3d render as caption
23
 
24
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6345bd89fe134dfd7a0dba40/jK75d8i1x5hAHSYSsJNBd.png" alt="Training configuration" style="max-height: 500px; width: auto;">
25
 
26
- Total step count is being 500 * 114 / 4 (4x GPU - batch size 1) = 14250
27
 
28
- Taking 37 hours currently if I don't terminate early
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
29
 
30
- Will save a checkpoint once every 25 epochs
31
 
32
  1 : https://youtu.be/bupRePUOA18
33
 
 
1
+ This is a training of a public LoRA style (4 seperate training each on 4x A6000).
2
 
3
  Experimenting captions vs non-captions. So we will see which yields best results for style training on FLUX.
4
 
 
23
 
24
  <img src="https://cdn-uploads.huggingface.co/production/uploads/6345bd89fe134dfd7a0dba40/jK75d8i1x5hAHSYSsJNBd.png" alt="Training configuration" style="max-height: 500px; width: auto;">
25
 
 
26
 
27
+ ## Inconsistent Dataset Training
28
+
29
+ This is the first training I made with the below dataset
30
+
31
+ [Inconsistent-Training-Dataset-Images-Grid.jpg](https://huggingface.co/MonsterMMORPG/3D-Cartoon-Style-FLUX/resolve/main/Inconsistent-Training-Dataset-Images-Grid.jpg)
32
+
33
+ When you pay attention to the grid image above shared, you will see that the dataset is not consistent
34
+
35
+ It has total 114 images
36
+
37
+ This training total step count was 500 * 114 / 4 (4x GPU - batch size 1) = 14250
38
+
39
+ It took like 37 hours on 4x RTX A6000 GPU with slow config - faster config would take like half
40
+
41
+ There were 2 trainings made with this dataset. Epoch 500 checkpoints are named as below
42
+
43
+ SECourses_Style_Inconsistent_DATASET_NO_Captions.safetensors
44
+ SECourses_Style_Inconsistent_DATASET_With_Captions.safetensors
45
+
46
+ Their checkpoints are saved in below folders
47
+
48
+ Training-Checkpoints-NO-Captions
49
+ Training-Checkpoints-With-Captions
50
+
51
+ Its grid results are shared below
52
+
53
+ https://huggingface.co/MonsterMMORPG/3D-Cartoon-Style-FLUX/resolve/main/Inconsistent-Training-Dataset-Results-Grid-26100x23700px.jpg
54
+
55
+ When you pay attention to above image you will see that it has inconsistent results
56
 
 
57
 
58
  1 : https://youtu.be/bupRePUOA18
59