All pull requests

Pull requests

• [BugFix] Update replay priorities when the Trainer has an optimization stepper

  BugFixCLA SignedTrainers
  #4716·sramireddy2 opened Nov 10, 2026Contributor·4 of 6 tasks·

  ·
  1

  1

• [BugFix] Let TensorDictMaxValueWriter extend with batches of trajectories

  BugFixCLA SignedReplayBuffers
  #4713·rishi02102017 opened Nov 9, 2026·4 tasks done·

  ·
  1

  3

• [BugFix] Let OneHot.project handle specs with more than one batch dim

  BugFixCLA Signed
  #4711·vishalhabib99 opened Nov 8, 2026·4 tasks done·

  ·
  1

  1

• [BugFix] Let BatchSubSampler draw the last sub-trajectory window

  BugFixCLA SignedTrainers
  #4709·jackiectl2 opened Nov 8, 2026Contributor·

  ·
  1

  1

• [Performance] Count trajectory splits in one pass

  BenchmarksCLA SignedCollectorsIntegrations/torch_geometricIntegrations
  #4707·Hang-get opened Nov 8, 2026·

  ·
  3

• [BugFix] FrameSkipTransform: stop at done and advance the state of stateless envs

  CLA SignedTransforms
  #4705·kaanutkuy opened Nov 7, 2026·6 tasks done·

  ·
  1

  1

• [BugFix] Preserve distributional DQN projection dtype

  BugFixCLA SignedObjectives
  #4697·davrens101 opened Nov 5, 2026·

  ·
  2

• [BugFix] Accept trainer-level network modules in on-policy, IQL and CQL configs

  CLA SignedTrainers
  #4690·sramireddy2 opened Nov 5, 2026Contributor·4 of 6 tasks·

  ·
  1

  1

• [BugFix] Fix optim step count and loss average in Trainer.optim_steps

  BugFixCLA SignedTrainers
  #4688·sramireddy2 opened Nov 5, 2026Contributor·4 of 6 tasks·

  ·
  1

  1

• [BugFix] Let H5 checkpoints follow TensorDict inference defaults

  BugFixCICLA SignedDocumentationReplayBuffers
  #4684·Z-Ghostshell opened Nov 4, 2026·8 tasks done·

  ·
  2

• [BugFix] Honor sum reduction in GRPOLoss token aggregation

  BugFixCLA Signedllm/Objectives
  #4682·Syycc-code opened Nov 4, 2026·7 tasks done·

  ·
  1

  3

• [BugFix] Reject //, ** and decimals in CountdownRewardParser.validate_expression

  BugFixCLA Signedllm/
  #4676·kaanutkuy opened Nov 3, 2026·6 tasks done·

  ·
  1

  6

• [LLM] Mask SFT reference losses to valid assistant tokens

  CLA Signedllm/Objectives
  #4673·davrens101 opened Nov 3, 2026·

  ·
  1

  1

• [Versioning] Delegate HDF5 batch metadata default

  CLA SignedReplayBuffersversioning
  #4671·bsprenger opened Nov 2, 2026Collaborator·6 tasks done·

  ·
  1

• [LLM] Detach clipped importance ratio in CISPOLoss to preserve token policy gradients

  CLA Signedllm/Objectives
  #4665·davrens101 opened Nov 2, 2026·

  ·
  1

  3

• [BugFix] Jointly broadcast TruncatedNormal parameters and bounds

  BugFixCLA SigneddistributionsIntegrations/torch_geometricIntegrationsModules
  #4662·DiadkoShmek opened Nov 2, 2026·4 of 5 tasks·

  ·
  1

  1

• [BugFix] Apply IndependentNormal location scaling during construction

  BugFixCLA SigneddistributionsIntegrations/torch_geometricIntegrationsModules
  #4660·DiadkoShmek opened Nov 1, 2026·6 tasks done·

  ·
  1

• [BugFix] Reduce multidimensional policy_version in StalenessAwareSampler

  BugFixCLA SignedReplayBuffers
  #4656·coder-jayp opened Oct 31, 2026Contributor·3 of 6 tasks·

  ·
  3

• [Feature] Add GSPOLoss

  CLA SignedDocumentationFeaturellm/Objectives
  #4653·arun-elr opened Oct 31, 2026Contributor·6 tasks done·

  ·
  1

  9

• [BugFix] Fix LeRobotPolicyWrapper.from_pretrained for current lerobot

  CICLA SignedIntegrations/torch_geometricIntegrationsModules
  #4651·arun-elr opened Oct 31, 2026Contributor·7 tasks done·

  ·
  1

  4

• [Quality] Compile truncated-normal policy construction

  ci/olddepsCLA SigneddistributionsIntegrations/torch_geometricIntegrationsModulesquality
  #4649·bsprenger opened Oct 30, 2026Collaborator·2/2·4 of 5 tasks·

  ·
  1

• Simplify LowLevelController specification

  BenchmarksCLA SignedIntegrations/torch_geometricIntegrationsModulesTrainersTransformstutorials/
  #4643·itwasabhi opened Oct 29, 2026Contributor·1 of 10 tasks·

  ·
  6

• [BugFix] Keep the masks of OneHot and MultiOneHot specs when indexing

  BugFixCLA Signed
  #4635·kjgggggg opened Oct 28, 2026·2 of 6 tasks·

  ·
  1

  2

• [BugFix] Keep Categorical and MultiCategorical masks through reshape, unflatten and to

  BugFixCLA Signed
  #4634·kjgggggg opened Oct 28, 2026·2 of 6 tasks·

  ·
  1

  2

• [Docs] Fix ModelBasedEnvBase constructor argument docs

  CLA SignedDocumentation
  #4630·mmporong opened Oct 27, 2026·2 of 6 tasks·

  ·
  3