sc20fg/base_model_custom_tokenizer

Full fine-tune vs google-t5/t5-base. Fine-tune: 100% of the weights changed, across 12 of 12 layers.

Tensors changed: Attention 146/146, Norms 62/62, other 49/49