thunderboltc/combined_mbart_sanlishTObangla_newsplit

Full fine-tune vs facebook/mbart-large-50-many-to-many-mmt. Fine-tune: 100% of the weights changed, across 12 of 12 layers.

Tensors changed: other 1/2, Embeddings 6/6, Norms 124/124, Attention 288/288, MLP 96/96

On the Hub · Downloads on Model Pulse