Commits

Branch selector

gumbel2

User selector

All users

All time

Commit history

Commits on May 27, 2023

• Add sampled top-k

  crowsonkb

  committed
  f36e6eeView commit details

  Copy full SHA for f36e6ee

  Browse repository at this point

• Fix overflow in fp16 when computing mean embedding during training

  crowsonkb

  committed
  47edacfView commit details

  Copy full SHA for 47edacf

  Browse repository at this point

Commits on May 26, 2023

• Exclude padding tokens from mean embedding when training reward models

  crowsonkb

  committed
  78ae67aView commit details

  Copy full SHA for 78ae67a

  Browse repository at this point

• Add LICENSE

  JD-P

  authored
  08f9b01View commit details

  Copy full SHA for 08f9b01

  Browse repository at this point

• Create reward_heads directory during training if not exists

  JD-P

  committed
  082f5b4View commit details

  Copy full SHA for 082f5b4

  Browse repository at this point

• Add ability to train reward head with zip upload within interface

  JD-P

  committed
  3f67795View commit details

  Copy full SHA for 3f67795

  Browse repository at this point

Commits on May 25, 2023

• Make the MLP actually an MLP

  JD-P

  committed
  b1de49dView commit details

  Copy full SHA for b1de49d

  Browse repository at this point

• Misc improvements and bug fixes

  Show description for 5c84f39
  JD-P

  committed
  5c84f39View commit details

  Copy full SHA for 5c84f39

  Browse repository at this point

• Initial commit

  JD-P

  committed
  5314b90View commit details

  Copy full SHA for 5314b90

  Browse repository at this point