{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:LNQ43R2QARWDLUX3NSLDWWVI3O","short_pith_number":"pith:LNQ43R2Q","schema_version":"1.0","canonical_sha256":"5b61cdc750046c35d2fb6c963b5aa8db8473142f53cf705929bf78fe7aa4698b","source":{"kind":"arxiv","id":"2407.06057","version":3},"attestation_state":"computed","paper":{"title":"Variational Best-of-N Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Afra Amini, Elliott Ash, Ryan Cotterell, Tim Vieira","submitted_at":"2024-07-08T15:59:44Z","abstract_excerpt":"Best-of-N (BoN) is a popular and effective algorithm for aligning language models to human preferences. The algorithm works as follows: at inference time, N samples are drawn from the language model, and the sample with the highest reward, as judged by a reward model, is returned as the output. Despite its effectiveness, BoN is computationally expensive; it reduces sampling throughput by a factor of N. To make BoN more efficient at inference time, one strategy is to fine-tune the language model to mimic what BoN does during inference. To achieve this, we derive the distribution induced by the "},"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":"2407.06057","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-07-08T15:59:44Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"ad66f7cffbf3813f95a991bf1d1270bee312f60c5457f6288c952b5acb17f271","abstract_canon_sha256":"b880367944008a8901ae5dbc28b132d235952eb607664cbe79d7f90641677645"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:23:54.803389Z","signature_b64":"2SendD0NYfFWaVgi8Oi3Z8O3urLJ+yGhmQlMwxI+tRhd9hn/ltt+nVrwKgY/lECpaOrc76o8az5yO777mxXsCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5b61cdc750046c35d2fb6c963b5aa8db8473142f53cf705929bf78fe7aa4698b","last_reissued_at":"2026-07-05T10:23:54.802671Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:23:54.802671Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Variational Best-of-N Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Afra Amini, Elliott Ash, Ryan Cotterell, Tim Vieira","submitted_at":"2024-07-08T15:59:44Z","abstract_excerpt":"Best-of-N (BoN) is a popular and effective algorithm for aligning language models to human preferences. The algorithm works as follows: at inference time, N samples are drawn from the language model, and the sample with the highest reward, as judged by a reward model, is returned as the output. Despite its effectiveness, BoN is computationally expensive; it reduces sampling throughput by a factor of N. To make BoN more efficient at inference time, one strategy is to fine-tune the language model to mimic what BoN does during inference. To achieve this, we derive the distribution induced by the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.06057","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/2407.06057/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":"2407.06057","created_at":"2026-07-05T10:23:54.802751+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.06057v3","created_at":"2026-07-05T10:23:54.802751+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.06057","created_at":"2026-07-05T10:23:54.802751+00:00"},{"alias_kind":"pith_short_12","alias_value":"LNQ43R2QARWD","created_at":"2026-07-05T10:23:54.802751+00:00"},{"alias_kind":"pith_short_16","alias_value":"LNQ43R2QARWDLUX3","created_at":"2026-07-05T10:23:54.802751+00:00"},{"alias_kind":"pith_short_8","alias_value":"LNQ43R2Q","created_at":"2026-07-05T10:23:54.802751+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.01490","citing_title":"Don't Let Gains FADE: Breaking Down Policy Gradient Weights in RL","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04559","citing_title":"Beyond Static Best-of-N: Bayesian List-wise Alignment for LLM-based Recommendation","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LNQ43R2QARWDLUX3NSLDWWVI3O","json":"https://pith.science/pith/LNQ43R2QARWDLUX3NSLDWWVI3O.json","graph_json":"https://pith.science/api/pith-number/LNQ43R2QARWDLUX3NSLDWWVI3O/graph.json","events_json":"https://pith.science/api/pith-number/LNQ43R2QARWDLUX3NSLDWWVI3O/events.json","paper":"https://pith.science/paper/LNQ43R2Q"},"agent_actions":{"view_html":"https://pith.science/pith/LNQ43R2QARWDLUX3NSLDWWVI3O","download_json":"https://pith.science/pith/LNQ43R2QARWDLUX3NSLDWWVI3O.json","view_paper":"https://pith.science/paper/LNQ43R2Q","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.06057&json=true","fetch_graph":"https://pith.science/api/pith-number/LNQ43R2QARWDLUX3NSLDWWVI3O/graph.json","fetch_events":"https://pith.science/api/pith-number/LNQ43R2QARWDLUX3NSLDWWVI3O/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LNQ43R2QARWDLUX3NSLDWWVI3O/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LNQ43R2QARWDLUX3NSLDWWVI3O/action/storage_attestation","attest_author":"https://pith.science/pith/LNQ43R2QARWDLUX3NSLDWWVI3O/action/author_attestation","sign_citation":"https://pith.science/pith/LNQ43R2QARWDLUX3NSLDWWVI3O/action/citation_signature","submit_replication":"https://pith.science/pith/LNQ43R2QARWDLUX3NSLDWWVI3O/action/replication_record"}},"created_at":"2026-07-05T10:23:54.802751+00:00","updated_at":"2026-07-05T10:23:54.802751+00:00"}