{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:YKJ74LWQWCXWM4AA4QMUAMCWPU","short_pith_number":"pith:YKJ74LWQ","schema_version":"1.0","canonical_sha256":"c293fe2ed0b0af667000e4194030567d3f29a2d129464e1344a84c240f116ae5","source":{"kind":"arxiv","id":"2506.06215","version":1},"attestation_state":"computed","paper":{"title":"Corrector Sampling in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Itai Gat, Neta Shaul, Uriel Singer, Yaron Lipman","submitted_at":"2025-06-06T16:21:35Z","abstract_excerpt":"Autoregressive language models accumulate errors due to their fixed, irrevocable left-to-right token generation. To address this, we propose a new sampling method called Resample-Previous-Tokens (RPT). RPT mitigates error accumulation by iteratively revisiting and potentially replacing tokens in a window of previously generated text. This method can be integrated into existing autoregressive models, preserving their next-token-prediction quality and speed. Fine-tuning a pretrained 8B parameter model with RPT for only 100B resulted in ~10% relative improvements on reasoning and coding benchmark"},"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":"2506.06215","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-06T16:21:35Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"53f956754988ed75d84e025f0e42dcff0c1e3908142c9617cfdefbba019a3b51","abstract_canon_sha256":"f131b390de6dece89d657b10c43bdfd52cecf6e367f36903437ec5179e2dd61b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:17:27.599812Z","signature_b64":"MFLSiGe1KYNZeHVH00/8+a619z8dARvaX7NPGAbWyLFP7iwQcrJD/1XeO9Ivp3rC/8oASCYA0gHipssGBFFuDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c293fe2ed0b0af667000e4194030567d3f29a2d129464e1344a84c240f116ae5","last_reissued_at":"2026-07-05T11:17:27.599325Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:17:27.599325Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Corrector Sampling in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Itai Gat, Neta Shaul, Uriel Singer, Yaron Lipman","submitted_at":"2025-06-06T16:21:35Z","abstract_excerpt":"Autoregressive language models accumulate errors due to their fixed, irrevocable left-to-right token generation. To address this, we propose a new sampling method called Resample-Previous-Tokens (RPT). RPT mitigates error accumulation by iteratively revisiting and potentially replacing tokens in a window of previously generated text. This method can be integrated into existing autoregressive models, preserving their next-token-prediction quality and speed. Fine-tuning a pretrained 8B parameter model with RPT for only 100B resulted in ~10% relative improvements on reasoning and coding benchmark"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.06215","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/2506.06215/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":"2506.06215","created_at":"2026-07-05T11:17:27.599385+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.06215v1","created_at":"2026-07-05T11:17:27.599385+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.06215","created_at":"2026-07-05T11:17:27.599385+00:00"},{"alias_kind":"pith_short_12","alias_value":"YKJ74LWQWCXW","created_at":"2026-07-05T11:17:27.599385+00:00"},{"alias_kind":"pith_short_16","alias_value":"YKJ74LWQWCXWM4AA","created_at":"2026-07-05T11:17:27.599385+00:00"},{"alias_kind":"pith_short_8","alias_value":"YKJ74LWQ","created_at":"2026-07-05T11:17:27.599385+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YKJ74LWQWCXWM4AA4QMUAMCWPU","json":"https://pith.science/pith/YKJ74LWQWCXWM4AA4QMUAMCWPU.json","graph_json":"https://pith.science/api/pith-number/YKJ74LWQWCXWM4AA4QMUAMCWPU/graph.json","events_json":"https://pith.science/api/pith-number/YKJ74LWQWCXWM4AA4QMUAMCWPU/events.json","paper":"https://pith.science/paper/YKJ74LWQ"},"agent_actions":{"view_html":"https://pith.science/pith/YKJ74LWQWCXWM4AA4QMUAMCWPU","download_json":"https://pith.science/pith/YKJ74LWQWCXWM4AA4QMUAMCWPU.json","view_paper":"https://pith.science/paper/YKJ74LWQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.06215&json=true","fetch_graph":"https://pith.science/api/pith-number/YKJ74LWQWCXWM4AA4QMUAMCWPU/graph.json","fetch_events":"https://pith.science/api/pith-number/YKJ74LWQWCXWM4AA4QMUAMCWPU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YKJ74LWQWCXWM4AA4QMUAMCWPU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YKJ74LWQWCXWM4AA4QMUAMCWPU/action/storage_attestation","attest_author":"https://pith.science/pith/YKJ74LWQWCXWM4AA4QMUAMCWPU/action/author_attestation","sign_citation":"https://pith.science/pith/YKJ74LWQWCXWM4AA4QMUAMCWPU/action/citation_signature","submit_replication":"https://pith.science/pith/YKJ74LWQWCXWM4AA4QMUAMCWPU/action/replication_record"}},"created_at":"2026-07-05T11:17:27.599385+00:00","updated_at":"2026-07-05T11:17:27.599385+00:00"}