Code

DPO#30

Code

Open

shahbuland wants to merge 30 commits intomainCarperAI/DRLX:mainfrom dpoCarperAI/DRLX:dpoCopy head branch name to clipboard

+1,050-401

Commits

Commits on Sep 17, 2023

• Add base trainer for any accelerate model

  shahbuland

  committed
  71c3014

  Copy full SHA for 71c3014

Commits on Sep 27, 2023

• Add PickaPic pipeline for DPO

  shahbuland

  committed
  af358b7

  Copy full SHA for af358b7

• add skeleton for DPO trainer

  shahbuland

  committed
  b140623

  Copy full SHA for b140623

Commits on Jan 23, 2024

• Pipeline for DPO

  

  Shahbuland Matiana

  committed
  704a85c

  Copy full SHA for 704a85c

• Allow for list of images instead of just list of np arrays for sample drawing

  

  Shahbuland Matiana

  committed
  68ec789

  Copy full SHA for 68ec789

• Add sampler for DPO

  

  Shahbuland Matiana

  committed
  e70e537

  Copy full SHA for e70e537

• Add method config for DPO

  

  Shahbuland Matiana

  committed
  2202d9f

  Copy full SHA for 2202d9f

• Add DPO trainer initial version

  

  Shahbuland Matiana

  committed
  9304bac

  Copy full SHA for 9304bac

Commits on Jan 24, 2024

• basic debugs

  

  shahbuland

  committed
  1752621

  Copy full SHA for 1752621

• Remove streaming

  

  shahbuland

  committed
  b11e873

  Copy full SHA for b11e873

• minor bug fixes

  

  shahbuland

  committed
  9201d6a

  Copy full SHA for 9201d6a

• Moved saving from DDPO trainer to base accelerate

  

  shahbuland

  committed
  35dd03d

  Copy full SHA for 35dd03d

Commits on Jan 25, 2024

• LoRA, refactorings, quick bug fixes

  

  shahbuland

  committed
  e16526f

  Copy full SHA for e16526f

• small bug fixes

  

  shahbuland

  committed
  14fe254

  Copy full SHA for 14fe254

• bug fixes

  

  shahbuland

  committed
  e121257

  Copy full SHA for e121257

• Fix import errors and checkpointing

  

  shahbuland

  committed
  6d9e03d

  Copy full SHA for 6d9e03d

Commits on Jan 26, 2024

• Add base model loss deviation to sampling as metric

  

  shahbuland

  committed
  c2350cb

  Copy full SHA for c2350cb

• Add base model loss deviation to trainer logging as metric

  

  shahbuland

  committed
  765b9f6

  Copy full SHA for 765b9f6

• Add non-lora training with memory saving options in config

  

  shahbuland

  committed
  74012cc

  Copy full SHA for 74012cc

Commits on Jan 28, 2024

• some refactorings to sampling, add rmsprop

  

  shahbuland

  committed
  ef91f92

  Copy full SHA for ef91f92

• Delete old DPO example, push new one

  

  shahbuland

  committed
  38847c5

  Copy full SHA for 38847c5

Commits on Feb 13, 2024

• Rename DPO2 to DPO

  

  shahbuland

  committed
  be05515

  Copy full SHA for be05515

• Move DPO and DDPO sampler to their own files for better organiation

  

  shahbuland

  committed
  e6023a3

  Copy full SHA for e6023a3

• prepare for adding SDXL

  

  shahbuland

  committed
  5253473

  Copy full SHA for 5253473

• Fix issue with modularizing samplers

  

  shahbuland

  committed
  54f6ec1

  Copy full SHA for 54f6ec1

• Add SDXL support and reorganize config for model

  

  shahbuland

  committed
  44f163d

  Copy full SHA for 44f163d

• Remove mandatory gradient clipping and fix model saving with new config

  

  shahbuland

  committed
  565efe6

  Copy full SHA for 565efe6

• Update dpo_pickapic.yml

  shahbuland

  authored
  4324932

  Copy full SHA for 4324932

• Update dpo_pickapic.yml

  shahbuland

  authored
  dde1265

  Copy full SHA for dde1265

• Update dpo_pickapic.yml

  shahbuland

  authored
  70f1827

  Copy full SHA for 70f1827