{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:BMTMCSSAWWNXG356H3UC6GUPVG","short_pith_number":"pith:BMTMCSSA","schema_version":"1.0","canonical_sha256":"0b26c14a40b59b736fbe3ee82f1a8fa9bdb0d5ac3f108140e925d453a576b92e","source":{"kind":"arxiv","id":"2403.17031","version":1},"attestation_state":"computed","paper":{"title":"The N+ Implementation Details of RLHF with PPO: A Case Study on TL;DR Summarization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Arian Hosseini, Kashif Rasul, Lewis Tunstall, Michael Noukhovitch, Shengyi Huang, Weixun Wang","submitted_at":"2024-03-24T02:59:27Z","abstract_excerpt":"This work is the first to openly reproduce the Reinforcement Learning from Human Feedback (RLHF) scaling behaviors reported in OpenAI's seminal TL;DR summarization work. We create an RLHF pipeline from scratch, enumerate over 20 key implementation details, and share key insights during the reproduction. Our RLHF-trained Pythia models demonstrate significant gains in response quality that scale with model size, with our 2.8B, 6.9B models outperforming OpenAI's released 1.3B checkpoint. We publicly release the trained model checkpoints and code to facilitate further research and accelerate progr"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2403.17031","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-03-24T02:59:27Z","cross_cats_sorted":[],"title_canon_sha256":"18306f6ff66983cbcf693227e189be7939b40e2211c68c5c270d4654d2be97c4","abstract_canon_sha256":"70a02a47cddd024f68ffb8636ef68ded555e6c6463a13c3b7b2e8a6c637e94e0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:00:44.139376Z","signature_b64":"fWwbMWZx8xswqNmAP3Rlzz2ybYcYlA+yGHhkuinbm60mWsFtNuMY7P5274VBDwQV+mQm6rHuX6jFxuAiOA5VAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0b26c14a40b59b736fbe3ee82f1a8fa9bdb0d5ac3f108140e925d453a576b92e","last_reissued_at":"2026-07-05T08:00:44.138900Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:00:44.138900Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The N+ Implementation Details of RLHF with PPO: A Case Study on TL;DR Summarization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Arian Hosseini, Kashif Rasul, Lewis Tunstall, Michael Noukhovitch, Shengyi Huang, Weixun Wang","submitted_at":"2024-03-24T02:59:27Z","abstract_excerpt":"This work is the first to openly reproduce the Reinforcement Learning from Human Feedback (RLHF) scaling behaviors reported in OpenAI's seminal TL;DR summarization work. We create an RLHF pipeline from scratch, enumerate over 20 key implementation details, and share key insights during the reproduction. Our RLHF-trained Pythia models demonstrate significant gains in response quality that scale with model size, with our 2.8B, 6.9B models outperforming OpenAI's released 1.3B checkpoint. We publicly release the trained model checkpoints and code to facilitate further research and accelerate progr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.17031","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2403.17031/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"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"},"aliases":[{"alias_kind":"arxiv","alias_value":"2403.17031","created_at":"2026-07-05T08:00:44.138958+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.17031v1","created_at":"2026-07-05T08:00:44.138958+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.17031","created_at":"2026-07-05T08:00:44.138958+00:00"},{"alias_kind":"pith_short_12","alias_value":"BMTMCSSAWWNX","created_at":"2026-07-05T08:00:44.138958+00:00"},{"alias_kind":"pith_short_16","alias_value":"BMTMCSSAWWNXG356","created_at":"2026-07-05T08:00:44.138958+00:00"},{"alias_kind":"pith_short_8","alias_value":"BMTMCSSA","created_at":"2026-07-05T08:00:44.138958+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21943","citing_title":"Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning","ref_index":79,"is_internal_anchor":false},{"citing_arxiv_id":"2509.20265","citing_title":"Failure Modes of Maximum Entropy RLHF","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2409.19256","citing_title":"HybridFlow: A Flexible and Efficient RLHF Framework","ref_index":34,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BMTMCSSAWWNXG356H3UC6GUPVG","json":"https://pith.science/pith/BMTMCSSAWWNXG356H3UC6GUPVG.json","graph_json":"https://pith.science/api/pith-number/BMTMCSSAWWNXG356H3UC6GUPVG/graph.json","events_json":"https://pith.science/api/pith-number/BMTMCSSAWWNXG356H3UC6GUPVG/events.json","paper":"https://pith.science/paper/BMTMCSSA"},"agent_actions":{"view_html":"https://pith.science/pith/BMTMCSSAWWNXG356H3UC6GUPVG","download_json":"https://pith.science/pith/BMTMCSSAWWNXG356H3UC6GUPVG.json","view_paper":"https://pith.science/paper/BMTMCSSA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.17031&json=true","fetch_graph":"https://pith.science/api/pith-number/BMTMCSSAWWNXG356H3UC6GUPVG/graph.json","fetch_events":"https://pith.science/api/pith-number/BMTMCSSAWWNXG356H3UC6GUPVG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BMTMCSSAWWNXG356H3UC6GUPVG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BMTMCSSAWWNXG356H3UC6GUPVG/action/storage_attestation","attest_author":"https://pith.science/pith/BMTMCSSAWWNXG356H3UC6GUPVG/action/author_attestation","sign_citation":"https://pith.science/pith/BMTMCSSAWWNXG356H3UC6GUPVG/action/citation_signature","submit_replication":"https://pith.science/pith/BMTMCSSAWWNXG356H3UC6GUPVG/action/replication_record"}},"created_at":"2026-07-05T08:00:44.138958+00:00","updated_at":"2026-07-05T08:00:44.138958+00:00"}