[WIP] Feature/rae#10

Draft

LouisRouss wants to merge 87 commits intomainfrom

Commits on Sep 10, 2025

Add RewardModel and PrefGRPORewardModel classes for reward computation
LouisRouss
committed

Commits on Sep 11, 2025

Refactor RewardModel and PrefGRPORewardModel to enhance image handling and reward computation - Allow batch processing with multiple prompts
LouisRouss
committed

Commits on Sep 16, 2025

Refactor training classes to use a common trainer and reorganize imports for clarity
LouisRouss
committed
Add GRPO support to Diffuser and Flow classes with new methods and utility updates
LouisRouss
committed
Enhance RewardModel and PrefGRPORewardModel with n_image_per_prompt support and advantage computation
LouisRouss
committed

Commits on Sep 19, 2025

Commits on Sep 21, 2025

Add clip in reward model
LouisRouss
committed

Commits on Sep 29, 2025

Implement DDT architecture and refactor modulation classes for enhanced flexibility and context handling
LouisRouss
committed

Commits on Oct 1, 2025

add dinoV3 and precompute functions
LouisRouss
committed

Commits on Oct 5, 2025

Refactor SD3TextEmbedder to improve type casting and add attention mask to embedding outputs + fix part of the code
LouisRouss
committed

Commits on Nov 11, 2025

Commits on Nov 17, 2025

Commits on Nov 20, 2025

feat(RAETrainer): add hinge GAN option for loss calculation
LouisRouss
committed