deepvk/USER2-small

Full fine-tune vs deepvk/RuModernBERT-small. Fine-tune: 100% of the weights changed, across 12 of 12 layers.

Tensors changed: Embeddings 2/2, Norms 24/24, Attention 24/24, MLP 24/24

1 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse