Code

RPO on multiple responses#311

Code

Draft

Davood-M wants to merge 20 commits intoNVIDIA:mainNVIDIA/NeMo-Aligner:mainfrom Davood-M:davidm/RPO_multi_respDavood-M/NeMo-Aligner:davidm/RPO_multi_respCopy head branch name to clipboard

+1,187-1

Commits

Commits on Jul 3, 2024

• fix log probs mismatch

  Show description for c713cec
  shengyangs

  committed
  c713cec

• fix log probs mismatch

  Show description for 0b11982
  shengyangs

  committed
  0b11982

• [pre-commit.ci] auto fixes from pre-commit.com hooks

  Show description for 78ef529
  pre-commit-ci[bot]

  committed
  78ef529

Commits on Jul 5, 2024

• enable log_prob_forward_micro_batch_size

  Show description for faee17d
  shengyangs

  committed
  faee17d

• resolve conficts

  Show description for 140860d
  shengyangs

  committed
  140860d

• [pre-commit.ci] auto fixes from pre-commit.com hooks

  Show description for dd6c271
  pre-commit-ci[bot]

  committed
  dd6c271

Commits on Jul 11, 2024

• add comments

  Show description for bfe3135
  shengyangs

  committed
  bfe3135

• Merge remote-tracking branch 'origin/ssy/fix-ref-policy' into ssy/fix-ref-policy

  Show description for 138c12d
  shengyangs

  committed
  138c12d

Commits on Sep 6, 2024

• added RPO for multiple responses (4)

  Show description for 0a81284
  Davood-M

  committed
  0a81284

• updated RPO loss

  Show description for cb6cbb7
  Davood-M

  committed
  cb6cbb7

• added metrics

  Show description for 2b34846
  

  David Mosallanezhad

  committed
  2b34846

Commits on Sep 10, 2024

• bug fixes

  Show description for d855f5f
  

  David Mosallanezhad

  committed
  d855f5f

Commits on Sep 12, 2024

• bug fix

  Show description for 780b134
  

  David Mosallanezhad

  committed
  780b134

Commits on Sep 16, 2024

• bugfix - numerical error

  Show description for 19d7882
  

  David Mosallanezhad

  committed
  19d7882

Commits on Sep 24, 2024

• updated RPO -> bug fixes

  Show description for 6b03c8c
  

  David Mosallanezhad

  committed
  6b03c8c

• updated

  Show description for a0764b2
  

  David Mosallanezhad

  committed
  a0764b2

• merged

  Show description for 655f2b6
  

  David Mosallanezhad

  committed
  655f2b6

• style fixes

  

  David Mosallanezhad

  committed
  539abdb

• style fixes

  

  David Mosallanezhad

  committed
  b2145fd

• [pre-commit.ci] auto fixes from pre-commit.com hooks

  Show description for bf90fdb
  pre-commit-ci[bot]

  committed
  bf90fdb