{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:IVASIGWSULKX4P74LQF2M3QW5I","short_pith_number":"pith:IVASIGWS","schema_version":"1.0","canonical_sha256":"4541241ad2a2d57e3ffc5c0ba66e16ea20f0edbb6f8d67f38dc61e9c595d93b4","source":{"kind":"arxiv","id":"2412.06000","version":1},"attestation_state":"computed","paper":{"title":"Does RLHF Scale? Exploring the Impacts From Data, Model, and Method","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Aohan Zeng, Hongning Wang, Jie Tang, Minlie Huang, Pengfan Du, Xiao Liu, Yilin Niu, Yuxiao Dong, Zhengxiao Du, Zhenyu Hou","submitted_at":"2024-12-08T17:19:48Z","abstract_excerpt":"This study explores the scaling properties of Reinforcement Learning from Human Feedback (RLHF) in Large Language Models (LLMs). Although RLHF is considered an important step in post-training of LLMs, its scaling potential is still largely unknown. We systematically analyze key components in the RLHF framework--model size, data composition, and inference budget--and their impacts on performance. Our findings show that increasing data diversity and volume improves reward model performance, helping process-supervision models scale better. For policy training, more response samples per prompt boo"},"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":"2412.06000","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-12-08T17:19:48Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"df18f33d165ba006f38fbea7bdf922a681da6e76af341c00dd7550cbc74b490f","abstract_canon_sha256":"5b1df4b5d67869fb2c6a68858dc6c12af14d3b86cfac48e49924a1069e78178c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:46:12.839838Z","signature_b64":"Fobb+Gu6IExApKGN/OkpcF6zE8TuEjYBjUrS9ieupWtWXx5RBUlaHW4EyErDCVc9wbePe6vlnuE0RujM/up5DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4541241ad2a2d57e3ffc5c0ba66e16ea20f0edbb6f8d67f38dc61e9c595d93b4","last_reissued_at":"2026-07-05T09:46:12.839372Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:46:12.839372Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Does RLHF Scale? Exploring the Impacts From Data, Model, and Method","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Aohan Zeng, Hongning Wang, Jie Tang, Minlie Huang, Pengfan Du, Xiao Liu, Yilin Niu, Yuxiao Dong, Zhengxiao Du, Zhenyu Hou","submitted_at":"2024-12-08T17:19:48Z","abstract_excerpt":"This study explores the scaling properties of Reinforcement Learning from Human Feedback (RLHF) in Large Language Models (LLMs). Although RLHF is considered an important step in post-training of LLMs, its scaling potential is still largely unknown. We systematically analyze key components in the RLHF framework--model size, data composition, and inference budget--and their impacts on performance. Our findings show that increasing data diversity and volume improves reward model performance, helping process-supervision models scale better. For policy training, more response samples per prompt boo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.06000","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/2412.06000/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":"2412.06000","created_at":"2026-07-05T09:46:12.839428+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.06000v1","created_at":"2026-07-05T09:46:12.839428+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.06000","created_at":"2026-07-05T09:46:12.839428+00:00"},{"alias_kind":"pith_short_12","alias_value":"IVASIGWSULKX","created_at":"2026-07-05T09:46:12.839428+00:00"},{"alias_kind":"pith_short_16","alias_value":"IVASIGWSULKX4P74","created_at":"2026-07-05T09:46:12.839428+00:00"},{"alias_kind":"pith_short_8","alias_value":"IVASIGWS","created_at":"2026-07-05T09:46:12.839428+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.05800","citing_title":"SALT: When More Rollouts Don't Help in Group-Based Policy Optimization and How to Make Them Matter","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2504.02181","citing_title":"A Survey of Scaling in Large Language Model Reasoning","ref_index":62,"is_internal_anchor":false},{"citing_arxiv_id":"2502.17419","citing_title":"From System 1 to System 2: A Survey of Reasoning Large Language Models","ref_index":299,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26326","citing_title":"Addressing Performance Saturation for LLM RL via Precise Entropy Curve Control","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26326","citing_title":"Addressing Performance Saturation for LLM RL via Precise Entropy Curve Control","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IVASIGWSULKX4P74LQF2M3QW5I","json":"https://pith.science/pith/IVASIGWSULKX4P74LQF2M3QW5I.json","graph_json":"https://pith.science/api/pith-number/IVASIGWSULKX4P74LQF2M3QW5I/graph.json","events_json":"https://pith.science/api/pith-number/IVASIGWSULKX4P74LQF2M3QW5I/events.json","paper":"https://pith.science/paper/IVASIGWS"},"agent_actions":{"view_html":"https://pith.science/pith/IVASIGWSULKX4P74LQF2M3QW5I","download_json":"https://pith.science/pith/IVASIGWSULKX4P74LQF2M3QW5I.json","view_paper":"https://pith.science/paper/IVASIGWS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.06000&json=true","fetch_graph":"https://pith.science/api/pith-number/IVASIGWSULKX4P74LQF2M3QW5I/graph.json","fetch_events":"https://pith.science/api/pith-number/IVASIGWSULKX4P74LQF2M3QW5I/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IVASIGWSULKX4P74LQF2M3QW5I/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IVASIGWSULKX4P74LQF2M3QW5I/action/storage_attestation","attest_author":"https://pith.science/pith/IVASIGWSULKX4P74LQF2M3QW5I/action/author_attestation","sign_citation":"https://pith.science/pith/IVASIGWSULKX4P74LQF2M3QW5I/action/citation_signature","submit_replication":"https://pith.science/pith/IVASIGWSULKX4P74LQF2M3QW5I/action/replication_record"}},"created_at":"2026-07-05T09:46:12.839428+00:00","updated_at":"2026-07-05T09:46:12.839428+00:00"}