-
Notifications
You must be signed in to change notification settings - Fork 2.8k
Pull requests: huggingface/trl
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Fix biased cross-rank aggregation of token-weighted metrics (GRPO, RLOO)
#6380
opened Jul 13, 2026 by
qgallouedec
Member
Loading…
Fix async GRPO clip-ratio extrema to global per-completion
#6379
opened Jul 13, 2026 by
qgallouedec
Member
Loading…
Simplify AsyncGRPOTrainer: drop redundant metric guards, merge train-begin callbacks
#6378
opened Jul 13, 2026 by
qgallouedec
Member
Loading…
Align
AsyncGRPOTrainer args handling with stable trainers
#6377
opened Jul 13, 2026 by
qgallouedec
Member
Loading…
8 tasks
Align
AsyncGRPOTrainer clip-ratio comment with GRPOTrainer
#6376
opened Jul 13, 2026 by
qgallouedec
Member
Loading…
8 tasks
Align
AsyncGRPOTrainer.processing_class docstring with GRPOTrainer
#6375
opened Jul 13, 2026 by
qgallouedec
Member
Loading…
8 tasks
Align
AsyncGRPOConfig.max_completion_length description with GRPOConfig
#6374
opened Jul 13, 2026 by
qgallouedec
Member
Loading…
8 tasks
Fix formatting of error message for loss_type
#6373
opened Jul 13, 2026 by
qgallouedec
Member
Loading…
8 tasks
Fix DPO/KTO
use_liger_kernel under DeepSpeed ZeRO-3
#6372
opened Jul 13, 2026 by
qgallouedec
Member
Loading…
Test all supported eval_dataset input types in core trainer evaluate() tests
#6371
opened Jul 13, 2026 by
albertvillanova
Member
Loading…
1 task
Fix DPO/KTO evaluate crash with a dict eval_dataset and precompute_ref_log_probs
#6370
opened Jul 13, 2026 by
albertvillanova
Member
Loading…
Stop overcounting dead groups and increase your accuracy
#6368
opened Jul 13, 2026 by
SushantDaga
Loading…
4 of 8 tasks
Avoid in-place mutation of input prompts in DPPO environments loop
#6366
opened Jul 12, 2026 by
albertvillanova
Member
Loading…
Avoid in-place mutation of input prompts in GRPO environments loop
#6364
opened Jul 12, 2026 by
albertvillanova
Member
Loading…
Fix corrupted reasoning blocks in Qwen3-family training chat templates
#6363
opened Jul 12, 2026 by
qgallouedec
Member
Loading…
Bump the actions group with 9 updates
dependencies
Pull requests that update a dependency file
github_actions
Pull requests that update GitHub Actions code
#6360
opened Jul 11, 2026 by
dependabot
Bot
Loading…
Truncate GOLD on-policy prompts before generation, keeping the prompt end
#6359
opened Jul 11, 2026 by
roycho96
Contributor
Loading…
3 tasks done
Add
trl.environments submodule with SandboxEnvironment
#6358
opened Jul 11, 2026 by
qgallouedec
Member
•
Draft
Fix EOS being masked when pad_token_id equals eos_token_id in GKD and GOLD
#6357
opened Jul 11, 2026 by
roycho96
Contributor
Loading…
3 tasks done
Force
use_reentrant=True for PEFT + ZeRO-3 + gradient checkpointing in all trainers
#6356
opened Jul 10, 2026 by
qgallouedec
Member
Loading…
Support iterable datasets in GRPO and RLOO
#6351
opened Jul 10, 2026 by
albertvillanova
Member
Loading…
Environment-owned dataset: make
train_dataset optional
#6349
opened Jul 10, 2026 by
qgallouedec
Member
Loading…
Preserve model forward signature in the chunked CE lm_head patch
#6342
opened Jul 9, 2026 by
sergiopaniego
Member
Loading…
3 of 6 tasks
Fix SDPO scale_rewards CLI parsing and warn on ignored feedback
#6341
opened Jul 9, 2026 by
sergiopaniego
Member
Loading…
3 of 6 tasks
Previous Next
ProTip!
Find all pull requests that aren't related to any open issues with -linked:issue.