Search results

Improve TRL import latency

Status: Open.

#5130 In huggingface/trl;

· albertvillanova opened on Feb 19, 2026

MGPO

Status: Open.

#5127 In huggingface/trl;

· damoonsh opened on Feb 19, 2026

Decouple rollout from vLLM [1/N]: Move rollout_func dispatch to top-level _generate_single_turn

Status: Open.

Feature

#5121 In huggingface/trl;

· albertvillanova opened on Feb 18, 2026

Preserve structured multimodal messages through rollout and generation pipeline

Status: Open.

Bug

#5120 In huggingface/trl;

· albertvillanova opened on Feb 18, 2026

Decouple inference backend from rollout & agent logic

Status: Open.

Feature

#5119 In huggingface/trl;

· albertvillanova opened on Feb 18, 2026

Training crash with "uncorrectable ECC error" and NCCL timeout?

Status: Open.

#5094 In huggingface/trl;

· 2402227288 opened on Feb 13, 2026

RFC: Defining the TRL Public API (towards v1.0)

Status: Open.

#5090 In huggingface/trl;

· qgallouedec opened on Feb 12, 2026

Source to GRPO online example

Status: Open.

#5033 In huggingface/trl;

· ttrung149 opened on Feb 9, 2026

Fine-Tuning gemma-3-4b-pt raises "ValueError: <code>token_type_ids</code> is required as a model input when training"

🏋 SFTRelated to SFTRelated to SFT

🐛 bugSomething isn't workingSomething isn't working

Status: Open.

#5032 In huggingface/trl;

· jesperschlegel opened on Feb 9, 2026

Maximum Likelihood Reinforcement Learning

Status: Open.

#5025 In huggingface/trl;

· catherinelee274 opened on Feb 9, 2026

DPPO - Divergence Proximal Policy Optimization

Status: Open.

#4998 In huggingface/trl;

· ErykCh opened on Feb 6, 2026

Support multi-turn training in GRPOTrainer

Status: Open.

Feature

#4974 In huggingface/trl;

· albertvillanova opened on Feb 5, 2026