{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:PEW4HTIJ6O6GYIC5M665N3X64C","short_pith_number":"pith:PEW4HTIJ","schema_version":"1.0","canonical_sha256":"792dc3cd09f3bc6c205d67bdd6eefee0aa9a9f34d9a8856a5ff9c8b59ff7997d","source":{"kind":"arxiv","id":"2405.19262","version":3},"attestation_state":"computed","paper":{"title":"Weak-to-Strong Search: Align Large Language Models via Searching over Small Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chao Yang, Jie Liu, Yu Qiao, Zhanhui Zhou, Zhichen Dong, Zhixuan Liu","submitted_at":"2024-05-29T16:55:32Z","abstract_excerpt":"Large language models are usually fine-tuned to align with human preferences. However, fine-tuning a large language model can be challenging. In this work, we introduce $\\textit{weak-to-strong search}$, framing the alignment of a large language model as a test-time greedy search to maximize the log-probability difference between small tuned and untuned models while sampling from the frozen large model. This method serves both as (1) a compute-efficient model up-scaling strategy that avoids directly tuning the large model and as (2) an instance of weak-to-strong generalization that enhances a s"},"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":"2405.19262","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-05-29T16:55:32Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"a0d1742a2645f447c0c0bdacdd4d3b240299aa0143dd9885a215f781682917de","abstract_canon_sha256":"414129eaec3205a5b78792f69d5ce6a70187b46b7de1c03d7aa64f109913f380"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:37:37.684577Z","signature_b64":"fS/hfDXXSQKmXpBfYOkVC2yuTSTfbWNtWgOTlWKHgKTy24k+B0ifWPOOBX+TAIcKcFp1Ob2l4XrQrZ/gN7knCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"792dc3cd09f3bc6c205d67bdd6eefee0aa9a9f34d9a8856a5ff9c8b59ff7997d","last_reissued_at":"2026-07-05T09:37:37.683780Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:37:37.683780Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Weak-to-Strong Search: Align Large Language Models via Searching over Small Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chao Yang, Jie Liu, Yu Qiao, Zhanhui Zhou, Zhichen Dong, Zhixuan Liu","submitted_at":"2024-05-29T16:55:32Z","abstract_excerpt":"Large language models are usually fine-tuned to align with human preferences. However, fine-tuning a large language model can be challenging. In this work, we introduce $\\textit{weak-to-strong search}$, framing the alignment of a large language model as a test-time greedy search to maximize the log-probability difference between small tuned and untuned models while sampling from the frozen large model. This method serves both as (1) a compute-efficient model up-scaling strategy that avoids directly tuning the large model and as (2) an instance of weak-to-strong generalization that enhances a s"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.19262","kind":"arxiv","version":3},"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/2405.19262/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":"2405.19262","created_at":"2026-07-05T09:37:37.683835+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.19262v3","created_at":"2026-07-05T09:37:37.683835+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.19262","created_at":"2026-07-05T09:37:37.683835+00:00"},{"alias_kind":"pith_short_12","alias_value":"PEW4HTIJ6O6G","created_at":"2026-07-05T09:37:37.683835+00:00"},{"alias_kind":"pith_short_16","alias_value":"PEW4HTIJ6O6GYIC5","created_at":"2026-07-05T09:37:37.683835+00:00"},{"alias_kind":"pith_short_8","alias_value":"PEW4HTIJ","created_at":"2026-07-05T09:37:37.683835+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.01682","citing_title":"Off-the-Shelf LLMs as Process Scorers: Training-Free Alternative to PRMs for Mathematical Reasoning","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01000","citing_title":"Trust Functions: Near-Lossless Weak-to-Strong Generalization by Learning When to Trust the Weak Teacher","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2502.01456","citing_title":"Process Reinforcement through Implicit Rewards","ref_index":81,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PEW4HTIJ6O6GYIC5M665N3X64C","json":"https://pith.science/pith/PEW4HTIJ6O6GYIC5M665N3X64C.json","graph_json":"https://pith.science/api/pith-number/PEW4HTIJ6O6GYIC5M665N3X64C/graph.json","events_json":"https://pith.science/api/pith-number/PEW4HTIJ6O6GYIC5M665N3X64C/events.json","paper":"https://pith.science/paper/PEW4HTIJ"},"agent_actions":{"view_html":"https://pith.science/pith/PEW4HTIJ6O6GYIC5M665N3X64C","download_json":"https://pith.science/pith/PEW4HTIJ6O6GYIC5M665N3X64C.json","view_paper":"https://pith.science/paper/PEW4HTIJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.19262&json=true","fetch_graph":"https://pith.science/api/pith-number/PEW4HTIJ6O6GYIC5M665N3X64C/graph.json","fetch_events":"https://pith.science/api/pith-number/PEW4HTIJ6O6GYIC5M665N3X64C/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PEW4HTIJ6O6GYIC5M665N3X64C/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PEW4HTIJ6O6GYIC5M665N3X64C/action/storage_attestation","attest_author":"https://pith.science/pith/PEW4HTIJ6O6GYIC5M665N3X64C/action/author_attestation","sign_citation":"https://pith.science/pith/PEW4HTIJ6O6GYIC5M665N3X64C/action/citation_signature","submit_replication":"https://pith.science/pith/PEW4HTIJ6O6GYIC5M665N3X64C/action/replication_record"}},"created_at":"2026-07-05T09:37:37.683835+00:00","updated_at":"2026-07-05T09:37:37.683835+00:00"}