{"paper":{"title":"OpenRLHF: An Easy-to-use, Scalable and High-performance RLHF Framework","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"OpenRLHF delivers a streamlined open-source framework for RLHF that trains models 1.22x to 1.68x faster while requiring far fewer lines of code.","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Bin Chen, Hao Chen, Haoran Wang, Haotian Xu, Jason Klein Liu, Jian Hu, Songlin Jiang, Weikai Fang, Wei Shen, Weixun Wang, Xianyu, Xibin Wu, Yiming Liu, Yu Cao, Zilin Zhu","submitted_at":"2024-05-20T01:04:40Z","abstract_excerpt":"Large Language Models (LLMs) fine-tuned via Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning with Verifiable Rewards (RLVR) significantly improve the alignment of human-AI values, further raising the upper bound of AI capabilities, particularly in reasoning-intensive, long-context Chain-of-Thought (CoT) tasks. However, existing frameworks commonly face challenges such as inference bottlenecks and complexity barriers, which restrict their accessibility to newcomers. To bridge this gap, we introduce \\textbf{OpenRLHF}, a user-friendly, scalable, and easy-to-learn open-"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experimental results show that OpenRLHF achieves superior training efficiency, with speedups ranging from 1.22x to 1.68x across different model sizes, compared to state-of-the-art frameworks. Additionally, it requires significantly fewer lines of code for implementation.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The reported speedups and code reductions are measured under fair, comparable conditions against the true state-of-the-art baselines, and the ease-of-use metric (lines of code) accurately reflects real-world implementation effort for typical users.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"OpenRLHF is a new open-source RLHF framework reporting 1.22x to 1.68x speedups and fewer lines of code than prior systems.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"OpenRLHF delivers a streamlined open-source framework for RLHF that trains models 1.22x to 1.68x faster while requiring far fewer lines of code.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ae0be8a697a3fc02167f98e44202e47dbd919e6c41b6a9fd0a9ea4d481e2e6df"},"source":{"id":"2405.11143","kind":"arxiv","version":6},"verdict":{"id":"edd6e13c-8c67-4d5f-93cf-1a2be2b4b07b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T03:24:18.968024Z","strongest_claim":"Experimental results show that OpenRLHF achieves superior training efficiency, with speedups ranging from 1.22x to 1.68x across different model sizes, compared to state-of-the-art frameworks. Additionally, it requires significantly fewer lines of code for implementation.","one_line_summary":"OpenRLHF is a new open-source RLHF framework reporting 1.22x to 1.68x speedups and fewer lines of code than prior systems.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The reported speedups and code reductions are measured under fair, comparable conditions against the true state-of-the-art baselines, and the ease-of-use metric (lines of code) accurately reflects real-world implementation effort for typical users.","pith_extraction_headline":"OpenRLHF delivers a streamlined open-source framework for RLHF that trains models 1.22x to 1.68x faster while requiring far fewer lines of code."},"references":{"count":30,"sample":[{"doi":"","year":2017,"title":"Deep reinforcement learning from human preferences.Advances in neural information processing systems, 30","work_id":"a63ca98b-8916-47bd-bd9b-d0c693024f4c","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Learning to summarize with human feedback","work_id":"4098eb4f-6e58-402f-9bde-2a625fa0675c","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","work_id":"e6b75ad5-2877-4168-97c8-710407094d20","ref_index":3,"cited_arxiv_id":"2501.12948","is_internal_anchor":true},{"doi":"","year":2025,"title":"Exploring data scaling trends and effects in reinforcement learning from human feedback","work_id":"b7419524-b2c2-49b3-9745-04133e8060b2","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","ref_index":5,"cited_arxiv_id":"2303.08774","is_internal_anchor":true}],"resolved_work":30,"snapshot_sha256":"ccd33241ac43c41a3506861d99c53b6a062aee36fbf96c8829ac817e291594f2","internal_anchors":10},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}