Repository Issue Activity (beta)

#7103 AsyncGRPO FSDP2 weight sync hangs: interleaved full_tensor() and is_checkpoint_format=True

#5191 Align trl skills CLI DX with HF CLI conventions

#7100 `num_tokens` overcounted under TP

#7093 CI emits 462 `transformers` `Trainer` log warnings: "The tokenizer has new PAD/BOS/EOS tokens that differ from the model config and generation config"

#7086 RLOO uses higher-variance KL Estimator and should use lower variance k3

#5127 MGPO

#5025 Maximum Likelihood Reinforcement Learning

#3549 Packing sequences for memory efficiency in GRPO and other preference learning implementations

#7071 SFTTrainer treats an empty config._name_or_path as a Hub repo id

#7083 Support precomputed reference log-probs in fused DPO training

#6833 RFC: Add single-image vision-language model support to A2POTrainer

#6832 RFC: Add end-to-end vision-language model support to SDPOTrainer

#6831 RFC: Support A2PO with DeepSpeed ZeRO-3 via an explicit reference model

#4339 Packing with VLMs

#7063 [Tracking] Move off liger-kernel and make the memory-efficient loss the default

#6970 Bug: pack_dataset BFD fails when all sequences are empty

#4184 [Feature] Add a bool enable/disable flag for `entropy logging`

#6221 [Feature] Reasoning reward utilities: group-relative length-scaled accuracy (GRPO-LEAD) + graduated format reward

#7012 PPOTrainer statistics average per-micro-batch means without weighting by micro-batch size

#7028 Does hf jobs uv run --image huggingface/trl actually use the preinstalled TRL?

#7024 MiniLLM: right-padding leaks into the reverse-KL advantage, and length normalization depends on batch padding

#7047 GRPO loss-path metrics average micro-batch entries, so uneven accumulation windows are weighted by their size

#6961 CI emits `transformers` warning log: Unrecognized keys in `rope_parameters` for 'rope_type'='default': {'full_attention', 'sliding_attention'}

#6857 Tiny test models are incompatible with Flash Attention (head_size=2)

#6877 Bug: unpair IterableDatasetDict preference datasets