deepset/roberta-base-squad2

Full fine-tune vs FacebookAI/roberta-base. Fine-tune: 100% of the weights changed, across 12 of 12 layers.

Tensors changed: Attention 96/96, other 48/48, Norms 48/48, Embeddings 5/6

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

On the Hub · Downloads on Model Pulse