VoxSign/Speech-to-Gloss
Full fine-tune vs google/flan-t5-base. Fine-tune: 90% of the weights changed, across 12 of 12 layers.
Tensors changed: Attention 145/146, Norms 58/62, other 72/73, lm_head 1/1
Full fine-tune vs google/flan-t5-base. Fine-tune: 90% of the weights changed, across 12 of 12 layers.
Tensors changed: Attention 145/146, Norms 58/62, other 72/73, lm_head 1/1