16 Pull requests merged by 5 people

• [Algorithm] Expert Iteration and SFT

  #3017 merged Jun 20, 2025

• [Feature] Add optional Explained Variance logging

  #3010 merged Jun 20, 2025

• [BugFix] ActionMask is compatible with composite action specs

  #3022 merged Jun 20, 2025

• [BugFix] Resolve deprec warning about warning

  #3024 merged Jun 20, 2025

• [Feature] Enabling worker level control on frames_per_batch

  #3020 merged Jun 20, 2025

• [BugFix] Fix wrong split_trajectories import

  #3023 merged Jun 20, 2025

• [BugFix] LLMCollector trajectory collection methods only returns 1 trajectory when multiple are completed at the same step

  #3018 merged Jun 18, 2025

• [BugFix] Fix cuda cache empty in GRPO scripts

  #3016 merged Jun 18, 2025

• [Feature] RayLLMCollector.sync_iter

  #3015 merged Jun 18, 2025

• [Feature] return_assistant_tokens_mask for SFT

  #3014 merged Jun 18, 2025

• [BugFix] Fix IFEval GRPO runs

  #3012 merged Jun 18, 2025

• [Algorithm] unify grpo sync/async implementations

  #3006 merged Jun 17, 2025

• [Doc] WeightUpdaterBase docs update after renaming

  #3007 merged Jun 17, 2025

• [Algorithm] Async GRPO

  #2997 merged Jun 16, 2025

• [BugFix] Fix deprecated list index

  #3005 merged Jun 16, 2025

• [BugFix] update_policy_weights_() with cudagraph

  #3003 merged Jun 16, 2025

3 Pull requests opened by 3 people

• [Feature, Example] A3C Atari Implementation for TorchRL

  #3001 opened Jun 15, 2025

• [Feature] Neptune logger

  #3008 opened Jun 17, 2025

• [Feature] Added EXP3 Scoring function in continuation with pr #2358

  #3013 opened Jun 18, 2025

3 Issues closed by 2 people

• [Feature Request] Having configurable `frames_per_batch_worker` on `_MultiDataCollector`

  #3019 closed Jun 20, 2025

• [BUG] MinariExperienceReplay imports non existent function from pytorch/rl

  #3021 closed Jun 20, 2025

• [BUG] vLLMWrapper prompt_logprobs kwarg should not be a boolean

  #3011 closed Jun 17, 2025

2 Issues opened by 2 people

• [Feature Request] Explained Variance

  #3009 opened Jun 17, 2025

• [BUG] CUDAgraph policy changes the learning process

  #3002 opened Jun 15, 2025

3 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• [Feature Request] Compressing data stored in the Replay Buffer

  #2983 commented on Jun 16, 2025 • 0 new comments

• [Feature Request] MCTS Issue tracker

  #2357 commented on Jun 17, 2025 • 0 new comments

• Minor fixes to wandb logger

  #2999 commented on Jun 16, 2025 • 0 new comments