tsmatz/roberta_qa_japanese
Full fine-tune vs rinna/japanese-roberta-base. Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: Embeddings 5/6, Norms 48/48, Attention 96/96, other 48/48
Full fine-tune vs rinna/japanese-roberta-base. Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: Embeddings 5/6, Norms 48/48, Attention 96/96, other 48/48