Skip to content

Release v0.4.6
Compare
Choose a tag to compare
@luozhouyang luozhouyang released this 29 Sep 03:16
· 29 commits to master since this release
v0.4.6
d8712a2

Changes:

  • Add Distiller for model distillation
  • Add support for difference padding strategies:
    • bucket: group sequence by length, and then padding sequence by max sequence length of the same batch
    • batch: padding sequence by max sequence length of this batch
    • fixed: padding sequence to a fixed max sequence length over all examples
Assets 2
Loading