{"paper":{"title":"RLHF Workflow: From Reward Modeling to Online RLHF","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Online iterative RLHF using proxy preference models from open-source datasets reaches state-of-the-art results on LLM chatbot benchmarks.","cross_cats":["cs.AI","cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Bo Pang, Caiming Xiong, Doyen Sahoo, HanZe Dong, Han Zhao, Haoxiang Wang, Nan Jiang, Tong Zhang, Wei Xiong, Yingbo Zhou","submitted_at":"2024-05-13T15:50:39Z","abstract_excerpt":"We present the workflow of Online Iterative Reinforcement Learning from Human Feedback (RLHF) in this technical report, which is widely reported to outperform its offline counterpart by a large margin in the recent large language model (LLM) literature. However, existing open-source RLHF projects are still largely confined to the offline learning setting. In this technical report, we aim to fill in this gap and provide a detailed recipe that is easy to reproduce for online iterative RLHF. In particular, since online human feedback is usually infeasible for open-source communities with limited "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We have shown that supervised fine-tuning (SFT) and iterative RLHF can obtain state-of-the-art performance with fully open-source datasets.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The proxy preference model built from open-source datasets approximates real human feedback closely enough that online RLHF updates remain beneficial rather than harmful.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"The paper supplies a complete open-source recipe for online iterative RLHF that uses proxy preference models and reaches competitive performance on AlpacaEval-2, Arena-Hard, and MT-Bench.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Online iterative RLHF using proxy preference models from open-source datasets reaches state-of-the-art results on LLM chatbot benchmarks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e6fa12b1e95f7cd16c20a0d0019b40f432fbbd80191bf69f844eaaae261c014a"},"source":{"id":"2405.07863","kind":"arxiv","version":3},"verdict":{"id":"8c2a5daa-ca4f-4c5d-bf1d-b112b840e968","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T16:43:20.020674Z","strongest_claim":"We have shown that supervised fine-tuning (SFT) and iterative RLHF can obtain state-of-the-art performance with fully open-source datasets.","one_line_summary":"The paper supplies a complete open-source recipe for online iterative RLHF that uses proxy preference models and reaches competitive performance on AlpacaEval-2, Arena-Hard, and MT-Bench.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The proxy preference model built from open-source datasets approximates real human feedback closely enough that online RLHF updates remain beneficial rather than harmful.","pith_extraction_headline":"Online iterative RLHF using proxy preference models from open-source datasets reaches state-of-the-art results on LLM chatbot benchmarks."},"references":{"count":23,"sample":[{"doi":"","year":null,"title":"Flat-leaf parsley is often preferred for its robust flavor","work_id":"3c2694e2-896c-47f0-8e99-d90dd293a9d5","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Garlic: Finely chopped or minced raw garlic gives a sharp, pungent taste to the Gremolata","work_id":"795ecb39-7a9d-45e9-87fe-a657ed7c6404","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"In some variations, you might also find:","work_id":"34e57393-c913-4f41-beef-c1879ce1f78d","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Lemon juice: This adds a little extra acidity and freshness to the mix. 5. Fresh basil or mint: These herbs can be used instead of, or in addition to, parsley, depending on the dish and personal prefe","work_id":"ce76a768-aba2-4aaa-b916-bcf98d0042c2","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Squats: Helps to strengthen the lower body, particularly the glutes, quadriceps, and hamstrings. 2. Push-ups: Improves upper body strength, targeting the chest, triceps, and shoulders. 3. Planks: Enga","work_id":"f910d4d1-1e52-4117-91f2-beab3e5ef5b6","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":23,"snapshot_sha256":"481a0d23db2a484444eb7d32b096bc5398924b4cdb6c5aa0656555238d98c04b","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"d12b6b4e4011789984be2d919d3be1f95b12751953a42594f86cb7a4e637d156"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}