Why we choose siglip-so400m-14-980-flash-attn2-navit?

Dear developers,

Your work is truly advanced and has given me a lot of inspiration. I am a beginner. Recently, while trying to train, I suddenly noticed a problem: I found that our ViT comes from **siglip-so400m-14-980-flash-attn2-navit**. I would like to ask, why can’t we just use the original **google/siglip-so400m-patch14-384**? Do we have to use **siglip-so400m-14-980-flash-attn2-navit**? If I want to use **google/siglip-so400m-patch14-384**, what changes would our code need to make?

I am very much looking forward to your reply.

Start by comparing the current siglip-so400m-14-980-flash-attn2-navit ViT choice with google/siglip-so400m-patch14-384 and identify where the model is configured during training. Determine the code and configuration changes needed for the alternate checkpoint, then verify that training runs successfully with it; the issue names no files or tests.

Written by the indexing model from the issue text.