Search results

Consider using pre-built Flash Attention kernels via kernels

Status: Open.

#1244 In OpenRLHF/OpenRLHF;

· sayakpaul opened on May 28, 2026

overlong_penalty should exclude tool response tokens from length calculation (use action_mask/action_ranges)

Status: Open.

#1243 In OpenRLHF/OpenRLHF;

· thevasudevgupta opened on May 26, 2026

ray error for multi-nodes training

Status: Open.

#1236 In OpenRLHF/OpenRLHF;

· RushDon opened on May 7, 2026

Torch 2.10's LRScheduler._update_lr breaks LORA

Status: Open.

#1225 In OpenRLHF/OpenRLHF;

· allen-dc opened on Apr 21, 2026

multimodal reward models

Status: Open.

#1224 In OpenRLHF/OpenRLHF;

· spirit-moon-fly opened on Apr 20, 2026

Sandboxed execution for multi-turn agent rollouts

Status: Open.

#1210 In OpenRLHF/OpenRLHF;

· congwang-mk opened on Apr 1, 2026

额外的奖励模型如何提升显卡利用率

Status: Open.

#1196 In OpenRLHF/OpenRLHF;

· pizts opened on Mar 19, 2026

For asynchronous training with vLLM 0.16.0+, please update to OpenRLHF 0.9.5

Status: Open.

#1193 In OpenRLHF/OpenRLHF;

· hijkzzz opened on Mar 5, 2026

DeepSpeed Off-loading Weights (not just Optimizer States)

Status: Open.

#1187 In OpenRLHF/OpenRLHF;

· jiosephlee opened on Feb 12, 2026

PPO "mini batches"

Status: Open.

#1185 In OpenRLHF/OpenRLHF;

· dawidm opened on Feb 10, 2026

Support for tool-calling

Status: Open.

#1183 In OpenRLHF/OpenRLHF;

· jiosephlee opened on Feb 6, 2026

Support Qwen3vl GRPO？

Status: Open.

#1175 In OpenRLHF/OpenRLHF;

· zhaoyuxin2 opened on Jan 15, 2026