{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:HSISQUDBVS73YMFMQD44O6HTWQ","short_pith_number":"pith:HSISQUDB","schema_version":"1.0","canonical_sha256":"3c91285061acbfbc30ac80f9c778f3b40254c135b80be8501940e1d16c9aacba","source":{"kind":"arxiv","id":"2508.05571","version":3},"attestation_state":"computed","paper":{"title":"iFairy: the First 2-bit Complex LLM with All Parameters in $\\{\\pm1, \\pm i\\}$","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Bokai Huang, Feiyu Wang, Guoan Wang, Rui Xu, Shengfan Wang, Shimao Chen, Tong Yang, Weitao Li, Yihao Zhang, Zihan Jiang","submitted_at":"2025-08-07T17:02:23Z","abstract_excerpt":"Quantization-Aware Training (QAT) integrates quantization into the training loop, enabling LLMs to learn robust low-bit representations, and is widely recognized as one of the most promising research directions. All current QAT research focuses on minimizing quantization error on full-precision models, where the full-precision accuracy acts as an upper bound (accuracy ceiling). No existing method has even attempted to surpass this ceiling. To break this ceiling, we propose a new paradigm: raising the ceiling (full-precision model), and then still quantizing it efficiently into 2 bits. We propo"},"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":"2508.05571","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-08-07T17:02:23Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"e8ca9b68b38ce5ad8c875cb7a0ac61c576c823d25d9dadbabab9d6c99302983d","abstract_canon_sha256":"15e2d04e9f906a0a95824490ec32e1053758484cd53c941e884cb56683d53672"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:54:50.423843Z","signature_b64":"ixjmVRJ0UM0zxpWQgiL7uXk+LFqP22fGPNY/cbYAY2ZJwV0NiTfuAng+0DlooXhkNrVLpw1an7LyIjXAs4jQCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3c91285061acbfbc30ac80f9c778f3b40254c135b80be8501940e1d16c9aacba","last_reissued_at":"2026-07-05T11:54:50.423380Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:54:50.423380Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"iFairy: the First 2-bit Complex LLM with All Parameters in $\\{\\pm1, \\pm i\\}$","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Bokai Huang, Feiyu Wang, Guoan Wang, Rui Xu, Shengfan Wang, Shimao Chen, Tong Yang, Weitao Li, Yihao Zhang, Zihan Jiang","submitted_at":"2025-08-07T17:02:23Z","abstract_excerpt":"Quantization-Aware Training (QAT) integrates quantization into the training loop, enabling LLMs to learn robust low-bit representations, and is widely recognized as one of the most promising research directions. All current QAT research focuses on minimizing quantization error on full-precision models, where the full-precision accuracy acts as an upper bound (accuracy ceiling). No existing method has even attempted to surpass this ceiling. To break this ceiling, we propose a new paradigm: raising the ceiling (full-precision model), and then still quantizing it efficiently into 2 bits. We propo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.05571","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/2508.05571/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":"2508.05571","created_at":"2026-07-05T11:54:50.423440+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.05571v3","created_at":"2026-07-05T11:54:50.423440+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.05571","created_at":"2026-07-05T11:54:50.423440+00:00"},{"alias_kind":"pith_short_12","alias_value":"HSISQUDBVS73","created_at":"2026-07-05T11:54:50.423440+00:00"},{"alias_kind":"pith_short_16","alias_value":"HSISQUDBVS73YMFM","created_at":"2026-07-05T11:54:50.423440+00:00"},{"alias_kind":"pith_short_8","alias_value":"HSISQUDB","created_at":"2026-07-05T11:54:50.423440+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.20913","citing_title":"FairyFuse: Multiplication-Free LLM Inference on CPUs via Fused Ternary Kernels","ref_index":35,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HSISQUDBVS73YMFMQD44O6HTWQ","json":"https://pith.science/pith/HSISQUDBVS73YMFMQD44O6HTWQ.json","graph_json":"https://pith.science/api/pith-number/HSISQUDBVS73YMFMQD44O6HTWQ/graph.json","events_json":"https://pith.science/api/pith-number/HSISQUDBVS73YMFMQD44O6HTWQ/events.json","paper":"https://pith.science/paper/HSISQUDB"},"agent_actions":{"view_html":"https://pith.science/pith/HSISQUDBVS73YMFMQD44O6HTWQ","download_json":"https://pith.science/pith/HSISQUDBVS73YMFMQD44O6HTWQ.json","view_paper":"https://pith.science/paper/HSISQUDB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.05571&json=true","fetch_graph":"https://pith.science/api/pith-number/HSISQUDBVS73YMFMQD44O6HTWQ/graph.json","fetch_events":"https://pith.science/api/pith-number/HSISQUDBVS73YMFMQD44O6HTWQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HSISQUDBVS73YMFMQD44O6HTWQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HSISQUDBVS73YMFMQD44O6HTWQ/action/storage_attestation","attest_author":"https://pith.science/pith/HSISQUDBVS73YMFMQD44O6HTWQ/action/author_attestation","sign_citation":"https://pith.science/pith/HSISQUDBVS73YMFMQD44O6HTWQ/action/citation_signature","submit_replication":"https://pith.science/pith/HSISQUDBVS73YMFMQD44O6HTWQ/action/replication_record"}},"created_at":"2026-07-05T11:54:50.423440+00:00","updated_at":"2026-07-05T11:54:50.423440+00:00"}