Eraly-ml/KazBERT
Full fine-tune vs google-bert/bert-base-uncased. Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: Embeddings 2/2, Attention 96/96, other 50/50
Full fine-tune vs google-bert/bert-base-uncased. Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: Embeddings 2/2, Attention 96/96, other 50/50