Sparse autoencoder for CLIP image embeddings. This was trained at 64x expansion.
See more at https://www.clipplayground.com
