Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on Mar 14, 2026

• feat(`grpo_trainer.py`): Variational Sequence-Level Soft Policy Optimization (VESPO) (#5199)

  casinca

  authored
  406d406View commit details

  Copy full SHA for 406d406

  Browse repository at this point

• Allow nullable logprobs in vLLM serve responses (#5203)

  Show description for d0ac7ef
  

  LeonEricsson

  and

  qgallouedec

  authored
  d0ac7efView commit details

  Copy full SHA for d0ac7ef

  Browse repository at this point

• Change default `vllm_mode` to `"colocate"` and add v0→v1 migration guide (#5255)

  Show description for c0eabc4
  

  qgallouedec

  and

  claude

  authored
  c0eabc4View commit details

  Copy full SHA for c0eabc4

  Browse repository at this point

Commits on Mar 13, 2026

• 35% faster packing + rename `bfd-requeue` to `bfd_split` (#5189)

  Show description for 6c0fccd
  

  mariosasko

  and

  qgallouedec

  authored
  6c0fccdView commit details

  Copy full SHA for 6c0fccd

  Browse repository at this point

• Raise ValueError for None train_dataset in experimental trainers (#5275)

  albertvillanova

  authored
  004ee19View commit details

  Copy full SHA for 004ee19

  Browse repository at this point

• Simplify get_train_dataloader in GRPO and RLOO (#5276)

  albertvillanova

  authored
  556646aView commit details

  Copy full SHA for 556646a

  Browse repository at this point

• Fix mm_token_type_ids silently dropped in DPO VLM training (#5279)

  albertvillanova

  authored
  e9c1821View commit details

  Copy full SHA for e9c1821

  Browse repository at this point

Commits on Mar 11, 2026

• Nemotron 3 examples added (#5272)

  Show description for 6648832
  

  sergiopaniego

  and

  claude

  authored
  6648832View commit details

  Copy full SHA for 6648832

  Browse repository at this point

• Fix title consistency from "Transformer Reinforcement Learning" to "Transformers Reinforcement Learning" (#5183)

  qgallouedec

  authored
  695fbe0View commit details

  Copy full SHA for 695fbe0

  Browse repository at this point

• Fix error message in OnlineDPO (#5237)

  qgallouedec

  authored
  f06f40bView commit details

  Copy full SHA for f06f40b

  Browse repository at this point

• Update TRL banner to support light/dark mode (#5270)

  qgallouedec

  authored
  bb8564aView commit details

  Copy full SHA for bb8564a

  Browse repository at this point

• Fix support for model_init_kwargs in GKD/GOLD when passed as CLI JSON string (#5266)

  albertvillanova

  authored
  11be774View commit details

  Copy full SHA for 11be774

  Browse repository at this point

• Align GOLDConfig docstrings for optional params with None default (#5261)

  albertvillanova

  authored
  e5ea2c4View commit details

  Copy full SHA for e5ea2c4

  Browse repository at this point

• Fix type for teacher_model_init_kwargs when passed as CLI JSON string (#5258)

  albertvillanova

  authored
  179108bView commit details

  Copy full SHA for 179108b

  Browse repository at this point

Commits on Mar 10, 2026

• [CPO/ORPO] Fix handling of different length chosen/rejected prompts. (#4639)

  Show description for b0a6bfd
  

  davmels

  and

  qgallouedec

  authored
  b0a6bfdView commit details

  Copy full SHA for b0a6bfd

  Browse repository at this point

• [GRPO/RLOO] Extract tokenize prompts from `_generate_single_turn` (#5240)

  Show description for a65c830
  

  qgallouedec

  and

  albertvillanova

  authored
  a65c830View commit details

  Copy full SHA for a65c830

  Browse repository at this point

• fix(ppo): add gradient_checkpointing_enable/disable to PolicyAndValueWrapper (#5245)

  Show description for 673550b
  

  zxshen

  and

  qgallouedec

  authored
  673550bView commit details

  Copy full SHA for 673550b

  Browse repository at this point

• Remove extra_fields dead code [3/N]: Remove extra_fields from GRPOTrainer._generate_single_turn return value (#5264)

  albertvillanova

  authored
  52547c5View commit details

  Copy full SHA for 52547c5

  Browse repository at this point

• Remove extra_fields dead code [2/N]: Remove extra_fields from VLLMGeneration.generate return value (#5263)

  albertvillanova

  authored
  5879856View commit details

  Copy full SHA for 5879856

  Browse repository at this point

• [GRPO/RLOO] Unify tokenization across all generation backends in `_generate_single_turn` (#5239)

  Show description for d5aaaea
  

  qgallouedec

  and

  albertvillanova

  authored
  d5aaaeaView commit details

  Copy full SHA for d5aaaea

  Browse repository at this point

• Remove extra_fields dead code [1/N]: Remove extra_fields handling from VLLMGeneration.generate (#5262)

  albertvillanova

  authored
  0064787View commit details

  Copy full SHA for 0064787

  Browse repository at this point

• Fix typo in docstring for teacher_model_init_kwargs (#5260)

  albertvillanova

  authored
  1c8dd9bView commit details

  Copy full SHA for 1c8dd9b

  Browse repository at this point

• Support JSON string parsing of teacher_model_init_kwargs in MiniLLMConfig (#5259)

  albertvillanova

  authored
  3fd447aView commit details

  Copy full SHA for 3fd447a

  Browse repository at this point

• Refactor CLI [14/N] : Remove TrainingArguments import from core trainers (#5257)

  albertvillanova

  authored
  41602bcView commit details

  Copy full SHA for 41602bc

  Browse repository at this point

• [GRPO/RLOO] Tokenize before vLLM generation call (#5238)

  Show description for b77f36f
  

  qgallouedec

  and

  albertvillanova

  authored
  b77f36fView commit details

  Copy full SHA for b77f36f

  Browse repository at this point

• Update vLLM version support to 0.17.0 (#5253)

  qgallouedec

  authored
  bd5307eView commit details

  Copy full SHA for bd5307e

  Browse repository at this point

• Update vLLM version support to 0.16.0 (#5252)

  qgallouedec

  authored
  7283ebeView commit details

  Copy full SHA for 7283ebe

  Browse repository at this point

• Sync entire prompt/completion token tensors before indexing (#5218)

  shawnghu

  authored
  fdb228cView commit details

  Copy full SHA for fdb228c

  Browse repository at this point

• Update vLLM version support to 0.15.0 (#5251)

  qgallouedec

  authored
  563591eView commit details

  Copy full SHA for 563591e

  Browse repository at this point

• [GRPO] In-place temperature scaling operation (#5254)

  winglian

  authored
  c49f375View commit details

  Copy full SHA for c49f375

  Browse repository at this point

• Fix `RewardFunc` type alias to reflect actual calling convention (#5246)

  Show description for e8c8f8d
  

  zxshen

  and

  qgallouedec

  authored
  e8c8f8dView commit details

  Copy full SHA for e8c8f8d

  Browse repository at this point

• Move `rollout_func` from `_generate_single_turn` to `_generate` (#5232)

  qgallouedec

  authored
  f3b3705View commit details

  Copy full SHA for f3b3705

  Browse repository at this point

Commits on Mar 9, 2026

• Add VLM support when passing raw token IDs to vLLM client (#5227)

  qgallouedec

  authored
  a262d9fView commit details

  Copy full SHA for a262d9f

  Browse repository at this point

• Deprecate `truncate_prompt_tokens` for vLLM 0.17.0 (#5248)

  Show description for 74f65c6
  

  winglian

  and

  qgallouedec

  authored
  74f65c6View commit details

  Copy full SHA for 74f65c6

  Browse repository at this point

• Add support for raw ids in `prompts` in vLLM client and server (#5225)

  qgallouedec

  authored
  9db3688View commit details

  Copy full SHA for 9db3688

  Browse repository at this point