{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:TGGUCAKQJC7WBTWHMTPQJXZW7W","short_pith_number":"pith:TGGUCAKQ","schema_version":"1.0","canonical_sha256":"998d41015048bf60cec764df04df36fdb5201f30eb3dd7bd1543a034e1424b52","source":{"kind":"arxiv","id":"2605.23081","version":1},"attestation_state":"computed","paper":{"title":"ThriftAttention: Selective Mixed Precision for Long-Context FP4 Attention","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Joe Sharratt","submitted_at":"2026-05-21T22:28:27Z","abstract_excerpt":"Efficient attention algorithms are critical to mitigate the quadratic cost of attention in long-context workloads. Prior work utilises block-scaled quantisation techniques on Blackwell GPUs to move attention computation to 4-bit precision to accelerate inference. However, these techniques result in significant quality degradation in long-context settings. We show that the output impact of quantisation error is highly non-uniform and increases with the importance of each query-key interaction, concentrating functionally relevant error in a small number of attention blocks that contain the most "},"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":"2605.23081","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-21T22:28:27Z","cross_cats_sorted":[],"title_canon_sha256":"7f4b9c5ead28e20788704872e8c7d17d98d97f1a2659a9fd8917415080d76f15","abstract_canon_sha256":"1159f242f1479a6e944a7c3d22742f30cb331a652bc2c5e923049df675588432"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-25T02:01:37.533898Z","signature_b64":"HJNbx6WarNazEW1EI1ntaBrgLv77ZP0d9AhxgbtOgVsOZfctigSkuzwUzBjv8QY79fTsjwsaROTosrkpTpDMAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"998d41015048bf60cec764df04df36fdb5201f30eb3dd7bd1543a034e1424b52","last_reissued_at":"2026-05-25T02:01:37.533384Z","signature_status":"signed_v1","first_computed_at":"2026-05-25T02:01:37.533384Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ThriftAttention: Selective Mixed Precision for Long-Context FP4 Attention","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Joe Sharratt","submitted_at":"2026-05-21T22:28:27Z","abstract_excerpt":"Efficient attention algorithms are critical to mitigate the quadratic cost of attention in long-context workloads. Prior work utilises block-scaled quantisation techniques on Blackwell GPUs to move attention computation to 4-bit precision to accelerate inference. However, these techniques result in significant quality degradation in long-context settings. We show that the output impact of quantisation error is highly non-uniform and increases with the importance of each query-key interaction, concentrating functionally relevant error in a small number of attention blocks that contain the most "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.23081","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/2605.23081/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":"2605.23081","created_at":"2026-05-25T02:01:37.533472+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.23081v1","created_at":"2026-05-25T02:01:37.533472+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.23081","created_at":"2026-05-25T02:01:37.533472+00:00"},{"alias_kind":"pith_short_12","alias_value":"TGGUCAKQJC7W","created_at":"2026-05-25T02:01:37.533472+00:00"},{"alias_kind":"pith_short_16","alias_value":"TGGUCAKQJC7WBTWH","created_at":"2026-05-25T02:01:37.533472+00:00"},{"alias_kind":"pith_short_8","alias_value":"TGGUCAKQ","created_at":"2026-05-25T02:01:37.533472+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/TGGUCAKQJC7WBTWHMTPQJXZW7W","json":"https://pith.science/pith/TGGUCAKQJC7WBTWHMTPQJXZW7W.json","graph_json":"https://pith.science/api/pith-number/TGGUCAKQJC7WBTWHMTPQJXZW7W/graph.json","events_json":"https://pith.science/api/pith-number/TGGUCAKQJC7WBTWHMTPQJXZW7W/events.json","paper":"https://pith.science/paper/TGGUCAKQ"},"agent_actions":{"view_html":"https://pith.science/pith/TGGUCAKQJC7WBTWHMTPQJXZW7W","download_json":"https://pith.science/pith/TGGUCAKQJC7WBTWHMTPQJXZW7W.json","view_paper":"https://pith.science/paper/TGGUCAKQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.23081&json=true","fetch_graph":"https://pith.science/api/pith-number/TGGUCAKQJC7WBTWHMTPQJXZW7W/graph.json","fetch_events":"https://pith.science/api/pith-number/TGGUCAKQJC7WBTWHMTPQJXZW7W/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TGGUCAKQJC7WBTWHMTPQJXZW7W/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TGGUCAKQJC7WBTWHMTPQJXZW7W/action/storage_attestation","attest_author":"https://pith.science/pith/TGGUCAKQJC7WBTWHMTPQJXZW7W/action/author_attestation","sign_citation":"https://pith.science/pith/TGGUCAKQJC7WBTWHMTPQJXZW7W/action/citation_signature","submit_replication":"https://pith.science/pith/TGGUCAKQJC7WBTWHMTPQJXZW7W/action/replication_record"}},"created_at":"2026-05-25T02:01:37.533472+00:00","updated_at":"2026-05-25T02:01:37.533472+00:00"}