add images and update paths
Damian Stewart
|
|
@ -25,7 +25,7 @@ The training proceeds rapidly over the first 50 steps, with the model quickly ge
|
|||
|
||||
| step 5 | step 17 | step 29 | step 41 | step 53 |
|
||||
|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|
|
||||
|  |  |  |  |  |
|
||||
|  |  |  |  |  |
|
||||
|
||||
(In all of these images the prompt is `ted bennett in a grey suit and tie with his arms spread out`.)
|
||||
|
||||
|
|
@ -33,13 +33,13 @@ The training proceeds rapidly over the first 50 steps, with the model quickly ge
|
|||
|
||||
| step 65 | step 77 | step 89 | step 101 | step 113 |
|
||||
|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|
|
||||
|  |  |  |  |  |
|
||||
|  |  |  |  |  |
|
||||
|
||||
**In the third phase**, the model is starting to overfit: edges start to get a bit weird, the saturation becomes stronger; everything feels a whole lot more *intense* somehow:
|
||||
|
||||
| step 125 | step 137 | step 149 | step 161 | step 173 |
|
||||
|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------|
|
||||
|  |  |  |  |  |
|
||||
|  |  |  |  |  |
|
||||
|
||||
When the model starts looking like this that means you've gone past the best point and it's time to stop. Normally you would detect this point by watching the outputs, but you can also watch the `loss/val` graph and try and stop the training before it starts to become unstable or trend upward.
|
||||
|
||||
|
|
|
|||
|
After Width: | Height: | Size: 462 KiB |
|
After Width: | Height: | Size: 76 KiB |
|
After Width: | Height: | Size: 95 KiB |
|
After Width: | Height: | Size: 75 KiB |
|
After Width: | Height: | Size: 73 KiB |
|
After Width: | Height: | Size: 79 KiB |
|
After Width: | Height: | Size: 68 KiB |
|
After Width: | Height: | Size: 68 KiB |
|
After Width: | Height: | Size: 66 KiB |
|
After Width: | Height: | Size: 68 KiB |
|
After Width: | Height: | Size: 75 KiB |
|
After Width: | Height: | Size: 72 KiB |
|
After Width: | Height: | Size: 82 KiB |
|
After Width: | Height: | Size: 93 KiB |
|
After Width: | Height: | Size: 83 KiB |
|
After Width: | Height: | Size: 76 KiB |
|
After Width: | Height: | Size: 172 KiB |
|
After Width: | Height: | Size: 183 KiB |