mtreviso's picture
Create README.md
eb6dda1
|
raw
history blame contribute delete
No virus
428 Bytes
metadata
license: afl-3.0
language: en
tags:
  - t5
datasets:
  - wikipedia

cT5-small left-to-right

Github: https://github.com/mtreviso/chunked-t5

This is a variant of cT5 that was trained with a left-to-right autoregressive decoding mask. As a consequence, it does not support parallel decoding, but it still predicts the end-of-chunk token </c> at the end of each chunk.