reazon-research/japanese-hubert-base-k2-rs35kh-bpe

Full fine-tune vs reazon-research/japanese-hubert-base-k2. Fine-tune: 100% of the weights changed, across 12 of 12 layers.

Tensors changed: Norms 54/54, Attention 96/96, MLP 48/48, Embeddings 3/3, other 9/9

On the Hub · Downloads on Model Pulse