About parameter sharing in t5-v1.1

xwwwww March 12, 2022, 7:41am 1

I’m kind of confused about why t5-v1_1 disable parameter sharing. What is this designed for?

Topic		Replies	Views
BigBirdPegasus with attention_type="original_full" vs T5 🤗Transformers	0	256	March 11, 2022
Gradient overflow when fine-tune t5 on CNN/DM dataset Beginners	5	1691	September 3, 2020
T5forConditionalGeneration Beginners	2	2283	September 15, 2020
Why is BART so much slower than T5? 🤗Transformers	0	351	September 17, 2021
Attention and hidden state details from t5 Beginners	0	198	November 12, 2020