Regarding Protein Sequence Embeddings Extracted by AMPLIFY

[luozhenjie1997]

Hi,
I'm currently using Amplify to extract protein sequence features and a diffusion model to generate latent space embeddings. But I've noticed that the embedding values output by Amplify are way too large in scale, which isn't suitable for diffusion modeling. Any suggestions on how I should fix this?
Thanks in advance.

[young-su-ko]

Probably the embedding from the last layer, Layer 31, can be used.