Update README.md

README.md

@@ -5,7 +5,13 @@ datasets:
 ---
 ## Check out my [blog](https://huggingface.co/blog/Fgdfgfthgr/typical-anime-image-style-dim)!
 
-# You can use 6 numbers to fully describe the style of an (anime) image!
+# Update 17/10/2025
+V4 released! This time instead of training a vision model from scratch, 
+it uses a simple mlp that takes the cls token from a [DINOv3](https://huggingface.co/collections/facebook/dinov3-68924841bd6b561778e31009) model to get the embedding.
+
+Far more accurate than the previous V3! You do need access to the DINOv3 with your HuggingFace token, though.
+
+# You can use 6/7 numbers to fully describe the style of an (anime) image!
 
 ## What's it and what could it do?
 Many diffusion models, though, choose to use artist tags to control the style of output images. 
@@ -33,16 +39,16 @@ With current version (v3):
 
 Training was done using [PyTorch Lightning](https://lightning.ai/).
 
-lr = 0.0001
+lr = 0.0005
 
-weight_decay = 0.0001
+weight_decay = 0.01
 
-[AdEMAMix](https://github.com/apple/ml-ademamix) optimizer
+AdamW optimizer
 
 ExponentialLR scheduler, with a gamma of 0.99, applied every epoch.
 
-Batch size of 1. [accumulate_grad_batches](https://lightning.ai/docs/pytorch/stable/advanced/training_tricks.html) of 16.
+Batch size of 9999 (so all data goes through the network at once).
 
-With every anchor image, 16 positive images and 16 negative images are used.
+With every anchor image, 4 positive images and 16 negative images are used.
 
-Trained for 15 epoches. On 2 A100 GPUs. A total of 3434 optimizer updates.
+Trained for 150 epoches. On a single RTX 3080 GPUs. A total of 150 optimizer updates.