Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on Jul 30, 2026

• Ignore benign bitsandbytes 4-bit blocksize alignment warning in CI (#6582)

  albertvillanova

  authored
  4b0c545View commit details

  Copy full SHA for 4b0c545

  Browse repository at this point

• Default `use_bias_correction_kl=True` in GRPOConfig (#6503)

  gowtham-sai-yadav

  authored
  185f738View commit details

  Copy full SHA for 185f738

  Browse repository at this point

Commits on Jul 29, 2026

• [DistillationTrainer refactor] Add `_get_last_hidden_state` (unwired) (#6525)

  qgallouedec

  authored
  1121382View commit details

  Copy full SHA for 1121382

  Browse repository at this point

• [DistillationTrainer refactor] Align the vLLM ctor block with GRPO (#6524)

  qgallouedec

  authored
  6f58edfView commit details

  Copy full SHA for 6f58edf

  Browse repository at this point

• [DistillationTrainer refactor] Align the generation_kwargs dict with GRPO (#6523)

  qgallouedec

  authored
  874dd31View commit details

  Copy full SHA for 874dd31

  Browse repository at this point

• [DistillationTrainer refactor] Switch generation to GRPO's stack; delete the buffer (#6522)

  qgallouedec

  authored
  0ad0f10View commit details

  Copy full SHA for 0ad0f10

  Browse repository at this point

• Fix test: don't call `add_response_schema` when the model already ships a response template (#6574)

  qgallouedec

  authored
  1af60b1View commit details

  Copy full SHA for 1af60b1

  Browse repository at this point

• Use `dtype` instead of `torch_dtype` (deprecated) (#6560)

  qgallouedec

  authored
  65a8dd4View commit details

  Copy full SHA for 65a8dd4

  Browse repository at this point

• Fix inaccuracies in the release instructions (#6566)

  albertvillanova

  authored
  de63151View commit details

  Copy full SHA for de63151

  Browse repository at this point

• Fix patch release instructions: Pull release branch instead of main (#6563)

  albertvillanova

  authored
  27da166View commit details

  Copy full SHA for 27da166

  Browse repository at this point

Commits on Jul 28, 2026

• Fix test_peft_with_quantization tests for bitsandbytes 0.50.0 (#6546)

  albertvillanova

  authored
  8d19a36View commit details

  Copy full SHA for 8d19a36

  Browse repository at this point

• Suppress deprecated torch.get_autocast_gpu_dtype() warning from mamba-ssm kernel (#6556)

  albertvillanova

  authored
  869b768View commit details

  Copy full SHA for 869b768

  Browse repository at this point

• Bump the actions group with 3 updates (#6540)

  Show description for 20cd679
  

  dependabot[bot]

  and

  albertvillanova

  authored
  20cd679View commit details

  Copy full SHA for 20cd679

  Browse repository at this point

• Point the opencode example at huggingface/OpenEnv (#6529)

  Show description for fd10a86
  

  sergiopaniego

  and

  qgallouedec

  authored
  fd10a86View commit details

  Copy full SHA for fd10a86

  Browse repository at this point

• Fix a bug in vLLM weight synchronization when `vllm_enable_sleep_mode=True` (#5313)

  Show description for c285dc1
  

  muupan

  and

  qgallouedec

  authored
  c285dc1View commit details

  Copy full SHA for c285dc1

  Browse repository at this point

• fix: avoid CopySlices when scaling policy logits (#6554)

  Show description for ba44e04
  

  DaoyuanLi2816

  and

  qgallouedec

  authored
  ba44e04View commit details

  Copy full SHA for ba44e04

  Browse repository at this point

Commits on Jul 27, 2026

• Revert xfail for NemotronH GRPO/RLOO tests now that transformers#47569 fixed the kernels bug (#6558)

  albertvillanova

  authored
  9a8a8d3View commit details

  Copy full SHA for 9a8a8d3

  Browse repository at this point

• Respect `TQDM_DISABLE` in DPO/KTO/BCO reference log-prob loops (#6507)

  qgallouedec

  authored
  493e12eView commit details

  Copy full SHA for 493e12e

  Browse repository at this point

• fix(ppo): exclude padding tokens from entropy calculation (#6121)

  Show description for 190c39c
  

  mukund1985

  and

  qgallouedec

  authored
  190c39cView commit details

  Copy full SHA for 190c39c

  Browse repository at this point

Commits on Jul 26, 2026

• Allow packing and padding-free on VLMs with text-only data (#6547)

  cris96spa

  authored
  a640cc5View commit details

  Copy full SHA for a640cc5

  Browse repository at this point

• added step time metric to AsyncGRPOTrainer (#6490)

  Show description for 7642f40
  

  AmineDiro

  and

  qgallouedec

  authored
  7642f40View commit details

  Copy full SHA for 7642f40

  Browse repository at this point

Commits on Jul 25, 2026

• Hotfix CI: Xfail NemotronH GRPO/RLOO tests against a broken transformers dev build (#6542)

  albertvillanova

  authored
  4708879View commit details

  Copy full SHA for 4708879

  Browse repository at this point

• Hotfix CI: Temporarily pin bitsandbytes < 0.50.0 (#6544)

  albertvillanova

  authored
  f4f9f76View commit details

  Copy full SHA for f4f9f76

  Browse repository at this point

• fix(grpo): correct DAPO/CISPO/VESPO loss normalization when steps_per_generation != gradient_accumulation_steps (#6024)

  Show description for e1b2e21
  

  3 peopleauthored
  e1b2e21View commit details

  Copy full SHA for e1b2e21

  Browse repository at this point

Commits on Jul 24, 2026

• Validate that reward functions return one reward per completion (#6534)

  gowtham-sai-yadav

  authored
  8697378View commit details

  Copy full SHA for 8697378

  Browse repository at this point

• Add Gemma4 response template (#6535)

  qgallouedec

  authored
  9d6c681View commit details

  Copy full SHA for 9d6c681

  Browse repository at this point

• Test train_dataset types at init for core trainers (#6494)

  albertvillanova

  authored
  dd94e42View commit details

  Copy full SHA for dd94e42

  Browse repository at this point

• [DistillationTrainer refactor] Add `_generate` + `_generate_and_score_completions` (unwired) (#6521)

  Show description for ee89b8e
  

  3 peopleauthored
  ee89b8eView commit details

  Copy full SHA for ee89b8e

  Browse repository at this point

• [DistillationTrainer refactor] Add `_generate_single_turn` (unwired) (#6513)

  qgallouedec

  authored
  82f6b2cView commit details

  Copy full SHA for 82f6b2c

  Browse repository at this point

• Reject unsupported train_dataset types in core trainers (#6493)

  albertvillanova

  authored
  86608c7View commit details

  Copy full SHA for 86608c7

  Browse repository at this point

• Fix CI quality: Fix copyright year in trl/experimental/async_grpo/openenv_harness.py (#6531)

  albertvillanova

  authored
  2dbef12View commit details

  Copy full SHA for 2dbef12

  Browse repository at this point

• Fix prepare_deepspeed crash with cpu offload optimizer (#5916)

  Show description for 752c7ff
  

  4 peopleauthored
  752c7ffView commit details

  Copy full SHA for 752c7ff

  Browse repository at this point

• [DistillationTrainer refactor] Add the ctor generation-supporting state (add-only) (#6512)

  qgallouedec

  authored
  a6fbeddView commit details

  Copy full SHA for a6fbedd

  Browse repository at this point

• Fix crash when training with Liger kernel on pre-Ampere GPUs (#6518)

  albertvillanova

  authored
  2095c30View commit details

  Copy full SHA for 2095c30

  Browse repository at this point

• [DistillationTrainer refactor] Add the shared generation config fields (add-only) (#6511)

  qgallouedec

  authored
  e48f090View commit details

  Copy full SHA for e48f090

  Browse repository at this point