jingyaogong/siglip2-base-p32-256-ve

Same weights, converted (fewer layers) vs google/siglip2-base-patch32-256. jingyaogong/siglip2-base-p32-256-ve matches part of google/siglip2-base-patch32-256 (208 of its 408 tensors; the rest of google/siglip2-base-patch32-256 has no counterpart). On the shared part: Same weights stored in another dtype (F32 → F16).

Tensors changed: Embeddings 0/3, Norms 0/52, MLP 0/52, Attention 0/100, other 0/1

On the Hub · Downloads on Model Pulse