{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:EEZ36GTL4WDJFIEFCLJDLC4NU6","short_pith_number":"pith:EEZ36GTL","schema_version":"1.0","canonical_sha256":"2133bf1a6be58692a08512d2358b8da7bf536f2b70a19d3879980975d9117e1b","source":{"kind":"arxiv","id":"2607.22694","version":1},"attestation_state":"computed","paper":{"title":"Bayesian Repetition Penalty: A Principled Adjacent-Conditional Framework for Reversing Attention Collapse in Autoregressive Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Bin Ma, Dong Li, Wenjie Fan","submitted_at":"2026-07-17T17:01:57Z","abstract_excerpt":"Attention collapse in autoregressive language models -- manifested as repetitive token loops where the model becomes trapped in self-reinforcing attractors -- is a persistent pathology that existing decoding-time heuristics fail to address at its root cause. We present a principled framework that penalises or compensates anomalous confidence arising from collapsed generation patterns, by comparing a token's observed frequency against its corpus prior through an adjacent-conditional probability construction. The resulting self-normalising penalty ratio $R=f(m,n,p)/f(np,n,p)$ requires no ad hoc "},"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":"2607.22694","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-07-17T17:01:57Z","cross_cats_sorted":[],"title_canon_sha256":"b0fbec8a01744a931127f071253ebcd1c49d0fea5cc313d710932c3df216180d","abstract_canon_sha256":"e5f9889235d91a5df658c2a37cdb758ea7ba76fdaa66c6bae888047f6d88f953"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-28T00:21:49.488942Z","signature_b64":"kdPLz3zAMeRuaTf1gBl7xBqg+3/HtjMNFM/EU/Q8VAmWcP6LBqWZLBDWQXWJc+mTnRowEsgLGc7yioEH5vxlAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2133bf1a6be58692a08512d2358b8da7bf536f2b70a19d3879980975d9117e1b","last_reissued_at":"2026-07-28T00:21:49.488039Z","signature_status":"signed_v1","first_computed_at":"2026-07-28T00:21:49.488039Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Bayesian Repetition Penalty: A Principled Adjacent-Conditional Framework for Reversing Attention Collapse in Autoregressive Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Bin Ma, Dong Li, Wenjie Fan","submitted_at":"2026-07-17T17:01:57Z","abstract_excerpt":"Attention collapse in autoregressive language models -- manifested as repetitive token loops where the model becomes trapped in self-reinforcing attractors -- is a persistent pathology that existing decoding-time heuristics fail to address at its root cause. We present a principled framework that penalises or compensates anomalous confidence arising from collapsed generation patterns, by comparing a token's observed frequency against its corpus prior through an adjacent-conditional probability construction. The resulting self-normalising penalty ratio $R=f(m,n,p)/f(np,n,p)$ requires no ad hoc "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.22694","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/2607.22694/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":"2607.22694","created_at":"2026-07-28T00:21:49.488518+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.22694v1","created_at":"2026-07-28T00:21:49.488518+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.22694","created_at":"2026-07-28T00:21:49.488518+00:00"},{"alias_kind":"pith_short_12","alias_value":"EEZ36GTL4WDJ","created_at":"2026-07-28T00:21:49.488518+00:00"},{"alias_kind":"pith_short_16","alias_value":"EEZ36GTL4WDJFIEF","created_at":"2026-07-28T00:21:49.488518+00:00"},{"alias_kind":"pith_short_8","alias_value":"EEZ36GTL","created_at":"2026-07-28T00:21:49.488518+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/EEZ36GTL4WDJFIEFCLJDLC4NU6","json":"https://pith.science/pith/EEZ36GTL4WDJFIEFCLJDLC4NU6.json","graph_json":"https://pith.science/api/pith-number/EEZ36GTL4WDJFIEFCLJDLC4NU6/graph.json","events_json":"https://pith.science/api/pith-number/EEZ36GTL4WDJFIEFCLJDLC4NU6/events.json","paper":"https://pith.science/paper/EEZ36GTL"},"agent_actions":{"view_html":"https://pith.science/pith/EEZ36GTL4WDJFIEFCLJDLC4NU6","download_json":"https://pith.science/pith/EEZ36GTL4WDJFIEFCLJDLC4NU6.json","view_paper":"https://pith.science/paper/EEZ36GTL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.22694&json=true","fetch_graph":"https://pith.science/api/pith-number/EEZ36GTL4WDJFIEFCLJDLC4NU6/graph.json","fetch_events":"https://pith.science/api/pith-number/EEZ36GTL4WDJFIEFCLJDLC4NU6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EEZ36GTL4WDJFIEFCLJDLC4NU6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EEZ36GTL4WDJFIEFCLJDLC4NU6/action/storage_attestation","attest_author":"https://pith.science/pith/EEZ36GTL4WDJFIEFCLJDLC4NU6/action/author_attestation","sign_citation":"https://pith.science/pith/EEZ36GTL4WDJFIEFCLJDLC4NU6/action/citation_signature","submit_replication":"https://pith.science/pith/EEZ36GTL4WDJFIEFCLJDLC4NU6/action/replication_record"}},"created_at":"2026-07-28T00:21:49.488518+00:00","updated_at":"2026-07-28T00:21:49.488518+00:00"}