Search results

AsyncGRPOTrainer: vision-language (*ForConditionalGeneration) checkpoints can't be trained (weight-sync key mismatch)

Status: Open.

#6028 In huggingface/trl;

· adithya-s-k opened on Jun 12, 2026

AsyncGRPOTrainer: environment_factory rewards are never passed to reward_funcs

Status: Open.

#6027 In huggingface/trl;

· adithya-s-k opened on Jun 12, 2026

Support training on Harbor agentic task suites via environment_factory

Status: Open.

#6017 In huggingface/trl;

· adithya-s-k opened on Jun 11, 2026

Add adapter-only vLLM LoRA sync for GRPO and AsyncGRPO

Status: Open.

#5975 In huggingface/trl;

· rycerzes opened on Jun 8, 2026

RFC: Rollout sources — a unified seam for environment & agent RL

Status: Open.

#5974 In huggingface/trl;

· adithya-s-k opened on Jun 8, 2026

RFC: Environment-owned reward

Status: Open.

#5912 In huggingface/trl;

· qgallouedec opened on Jun 2, 2026

RFC: Agent RL training with environment: environment-owned dataset, if any

Status: Open.

#5903 In huggingface/trl;

· qgallouedec opened on Jun 1, 2026

bug(rloo): KL penalty uses first-order log ratio instead of Schulman approximation (inconsistent with GRPOTrainer)

Status: Open.

#5889 In huggingface/trl;

· Sumu004 opened on May 30, 2026

RDPO support in GRPOTRainer

Status: Open.

#5886 In huggingface/trl;

· JishnuKR opened on May 29, 2026

AsyncGRPO: make NCCL weight sync robust on PCIe-only GPUs (auto fallback from P2P/SHM)

Status: Open.

#5865 In huggingface/trl;

· rycerzes opened on May 27, 2026

[AsyncGRPO] Loading from checkpoint recomputes all vLLM past requests

Status: Open.

#5863 In huggingface/trl;

· ggcr opened on May 27, 2026

[AsyncGRPO] aiohttp limits to 100 reqs when max_inflight_tasks > 100

Status: Open.

#5847 In huggingface/trl;

· ggcr opened on May 26, 2026