sergeyzh/BERTA

Full fine-tune vs sergeyzh/LaBSE-ru-turbo. Fine-tune: 100% of the weights changed, across 12 of 12 layers.

Tensors changed: Embeddings 5/5, Norms 48/48, Attention 96/96, other 48/50

2 derivatives checked

• sergeyzh/rubert-base-retriever: Full fine-tune
• sergeyzh/BERTA-uncased: Full fine-tune

On the Hub · Downloads on Model Pulse