Search results

如何理解critic scheduler initialization depends on max_step, so we have to init critic after actor

Status: Open.

#1121 In OpenRLHF/OpenRLHF;

· Ind1x1 opened on Aug 26, 2025

Training GRPO, the loss suddenly becomes nan and the reward becomes 0

Status: Open.

#1120 In OpenRLHF/OpenRLHF;

· baowj-678 opened on Aug 26, 2025

Token id out of vocabulary

Status: Open.

#1117 In OpenRLHF/OpenRLHF;

· qianxiao1111 opened on Aug 25, 2025

Proposal: Fully Asynchronous RL Pipeline Support in OpenRLHF

Status: Open.

#1116 In OpenRLHF/OpenRLHF;

· jiaxilyu opened on Aug 24, 2025

DeepSpeed AutoTP with Hybrid Engine

Status: Open.

#1110 In OpenRLHF/OpenRLHF;

· hijkzzz opened on Aug 21, 2025

RL Reward Instability and Reproducibility Issues Tied to vLLM Version

Status: Open.

#1108 In OpenRLHF/OpenRLHF;

· RowitZou opened on Aug 18, 2025

wandb中记录的train/generated_samples，sample和reward不是对应的吗

Status: Open.

#1107 In OpenRLHF/OpenRLHF;

· YSQ-boop opened on Aug 17, 2025

agent_func、reward_func时--normalization_reward参数不起作用，对吗

Status: Open.

#1106 In OpenRLHF/OpenRLHF;

· YSQ-boop opened on Aug 16, 2025

[BUG] Cannot reload the dataloader state from checkpoints in ppo training

Status: Open.

#1103 In OpenRLHF/OpenRLHF;

· cafeii opened on Aug 11, 2025

Support for NVIDIA Nsight Systems

Status: Open.

#1099 In OpenRLHF/OpenRLHF;

· Marshall-Ge opened on Aug 6, 2025

Are you considering supporting the FR3E algorithm？

Status: Open.

#1093 In OpenRLHF/OpenRLHF;

· AXy1527 opened on Jul 20, 2025

Release artifacts (models, dataset) on Hugging Face

Status: Open.

#1087 In OpenRLHF/OpenRLHF;

· NielsRogge opened on Jul 16, 2025