{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:6YEEQNVRVEGJDXCVNRHZAVJH2M","short_pith_number":"pith:6YEEQNVR","schema_version":"1.0","canonical_sha256":"f6084836b1a90c91dc556c4f905527d33a9bd06ca8b363e326fa0b57e624629b","source":{"kind":"arxiv","id":"2603.00040","version":3},"attestation_state":"computed","paper":{"title":"Attn-QAT: 4-Bit Attention With Quantization-Aware Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chengquan Jiang, Hao Zhang, Matthew Noto, Peiyuan Zhang, Wei Zhou, Wenxuan Tan, Will Lin","submitted_at":"2026-02-09T04:46:21Z","abstract_excerpt":"Achieving reliable 4-bit attention is a prerequisite for end-to-end FP4 computation on emerging FP4-capable GPUs, yet attention remains the main obstacle due to FP4's tiny dynamic range and attention's heavy-tailed activations. This paper presents the first systematic study of 4-bit quantization-aware training (QAT) for attention. We find ``drop-in'' QAT -- naively combining an FP4 forward pass with high-precision Flash Attention (FA)-style backward pass -- leads to training instability. We identify two key principles for stable FP4 attention: (1) matching low-precision recomputation of attent"},"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":"2603.00040","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-09T04:46:21Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"a62d6e6153932bad584f56ca66c3a8f819f08072c396fa552ba1078a4dc4a73b","abstract_canon_sha256":"235e5a91709bbf4c508a76670621a8b44961b376517e0e9433a6ab508234b56b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-11T01:24:39.225242Z","signature_b64":"f+gD1X5O/99ksb49IKixj06RvduG97DotTQgUpTjpeadq6UkLqLZOzZziqeIqmF5tUMYgoyS4f4G4LYgIy3xCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f6084836b1a90c91dc556c4f905527d33a9bd06ca8b363e326fa0b57e624629b","last_reissued_at":"2026-08-11T01:24:39.222650Z","signature_status":"signed_v1","first_computed_at":"2026-08-11T01:24:39.222650Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Attn-QAT: 4-Bit Attention With Quantization-Aware Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chengquan Jiang, Hao Zhang, Matthew Noto, Peiyuan Zhang, Wei Zhou, Wenxuan Tan, Will Lin","submitted_at":"2026-02-09T04:46:21Z","abstract_excerpt":"Achieving reliable 4-bit attention is a prerequisite for end-to-end FP4 computation on emerging FP4-capable GPUs, yet attention remains the main obstacle due to FP4's tiny dynamic range and attention's heavy-tailed activations. This paper presents the first systematic study of 4-bit quantization-aware training (QAT) for attention. We find ``drop-in'' QAT -- naively combining an FP4 forward pass with high-precision Flash Attention (FA)-style backward pass -- leads to training instability. We identify two key principles for stable FP4 attention: (1) matching low-precision recomputation of attent"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.00040","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/2603.00040/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":"2603.00040","created_at":"2026-08-11T01:24:39.223431+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.00040v3","created_at":"2026-08-11T01:24:39.223431+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.00040","created_at":"2026-08-11T01:24:39.223431+00:00"},{"alias_kind":"pith_short_12","alias_value":"6YEEQNVRVEGJ","created_at":"2026-08-11T01:24:39.223431+00:00"},{"alias_kind":"pith_short_16","alias_value":"6YEEQNVRVEGJDXCV","created_at":"2026-08-11T01:24:39.223431+00:00"},{"alias_kind":"pith_short_8","alias_value":"6YEEQNVR","created_at":"2026-08-11T01:24:39.223431+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/6YEEQNVRVEGJDXCVNRHZAVJH2M","json":"https://pith.science/pith/6YEEQNVRVEGJDXCVNRHZAVJH2M.json","graph_json":"https://pith.science/api/pith-number/6YEEQNVRVEGJDXCVNRHZAVJH2M/graph.json","events_json":"https://pith.science/api/pith-number/6YEEQNVRVEGJDXCVNRHZAVJH2M/events.json","paper":"https://pith.science/paper/6YEEQNVR"},"agent_actions":{"view_html":"https://pith.science/pith/6YEEQNVRVEGJDXCVNRHZAVJH2M","download_json":"https://pith.science/pith/6YEEQNVRVEGJDXCVNRHZAVJH2M.json","view_paper":"https://pith.science/paper/6YEEQNVR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.00040&json=true","fetch_graph":"https://pith.science/api/pith-number/6YEEQNVRVEGJDXCVNRHZAVJH2M/graph.json","fetch_events":"https://pith.science/api/pith-number/6YEEQNVRVEGJDXCVNRHZAVJH2M/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6YEEQNVRVEGJDXCVNRHZAVJH2M/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6YEEQNVRVEGJDXCVNRHZAVJH2M/action/storage_attestation","attest_author":"https://pith.science/pith/6YEEQNVRVEGJDXCVNRHZAVJH2M/action/author_attestation","sign_citation":"https://pith.science/pith/6YEEQNVRVEGJDXCVNRHZAVJH2M/action/citation_signature","submit_replication":"https://pith.science/pith/6YEEQNVRVEGJDXCVNRHZAVJH2M/action/replication_record"}},"created_at":"2026-08-11T01:24:39.223431+00:00","updated_at":"2026-08-11T01:24:39.223431+00:00"}