{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:ARICQWDRUOPEKLEW7L7XMJKPJ5","short_pith_number":"pith:ARICQWDR","schema_version":"1.0","canonical_sha256":"0450285871a39e452c96faff76254f4f4beade24dfc81d58080f8c04b60342c0","source":{"kind":"arxiv","id":"2012.15701","version":2},"attestation_state":"computed","paper":{"title":"BinaryBERT: Pushing the Limit of BERT Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Haoli Bai, Irwin King, Jing Jin, Lifeng Shang, Lu Hou, Michael Lyu, Qun Liu, Wei Zhang, Xin Jiang","submitted_at":"2020-12-31T16:34:54Z","abstract_excerpt":"The rapid development of large pre-trained language models has greatly increased the demand for model compression techniques, among which quantization is a popular solution. In this paper, we propose BinaryBERT, which pushes BERT quantization to the limit by weight binarization. We find that a binary BERT is hard to be trained directly than a ternary counterpart due to its complex and irregular loss landscape. Therefore, we propose ternary weight splitting, which initializes BinaryBERT by equivalently splitting from a half-sized ternary network. The binary model thus inherits the good performa"},"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":"2012.15701","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-12-31T16:34:54Z","cross_cats_sorted":[],"title_canon_sha256":"ee296fdf93d2baa61fbf9888d0a56a079704cbec7ea7eddf09ce86a309b12c33","abstract_canon_sha256":"bc9eaab8883742302ddfc14bcb0e8e40ea8a92615ddbefc6509030e2f15adbaa"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:59:54.975574Z","signature_b64":"hR7aoZ2fNwl141QiFqwBuyjnJ+PtXo++61EgqnqPDX3d+zG6XyRI4M8OS7kOG3OhZCTo39Xx8hz9/2udRxt2AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0450285871a39e452c96faff76254f4f4beade24dfc81d58080f8c04b60342c0","last_reissued_at":"2026-07-05T02:59:54.975170Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:59:54.975170Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"BinaryBERT: Pushing the Limit of BERT Quantization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Haoli Bai, Irwin King, Jing Jin, Lifeng Shang, Lu Hou, Michael Lyu, Qun Liu, Wei Zhang, Xin Jiang","submitted_at":"2020-12-31T16:34:54Z","abstract_excerpt":"The rapid development of large pre-trained language models has greatly increased the demand for model compression techniques, among which quantization is a popular solution. In this paper, we propose BinaryBERT, which pushes BERT quantization to the limit by weight binarization. We find that a binary BERT is hard to be trained directly than a ternary counterpart due to its complex and irregular loss landscape. Therefore, we propose ternary weight splitting, which initializes BinaryBERT by equivalently splitting from a half-sized ternary network. The binary model thus inherits the good performa"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2012.15701","kind":"arxiv","version":2},"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/2012.15701/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":"2012.15701","created_at":"2026-07-05T02:59:54.975227+00:00"},{"alias_kind":"arxiv_version","alias_value":"2012.15701v2","created_at":"2026-07-05T02:59:54.975227+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2012.15701","created_at":"2026-07-05T02:59:54.975227+00:00"},{"alias_kind":"pith_short_12","alias_value":"ARICQWDRUOPE","created_at":"2026-07-05T02:59:54.975227+00:00"},{"alias_kind":"pith_short_16","alias_value":"ARICQWDRUOPEKLEW","created_at":"2026-07-05T02:59:54.975227+00:00"},{"alias_kind":"pith_short_8","alias_value":"ARICQWDR","created_at":"2026-07-05T02:59:54.975227+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.27759","citing_title":"Layerwise Progressive Freezing: A Training Scaffold for Depth-Scalable Binary Networks","ref_index":71,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10989","citing_title":"SURGE: Surrogate Gradient Adaptation in Binary Neural Networks","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2208.07339","citing_title":"LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale","ref_index":119,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10989","citing_title":"SURGE: Surrogate Gradient Adaptation in Binary Neural Networks","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2604.04523","citing_title":"LOCALUT: Harnessing Capacity-Computation Tradeoffs for LUT-Based Inference in DRAM-PIM","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ARICQWDRUOPEKLEW7L7XMJKPJ5","json":"https://pith.science/pith/ARICQWDRUOPEKLEW7L7XMJKPJ5.json","graph_json":"https://pith.science/api/pith-number/ARICQWDRUOPEKLEW7L7XMJKPJ5/graph.json","events_json":"https://pith.science/api/pith-number/ARICQWDRUOPEKLEW7L7XMJKPJ5/events.json","paper":"https://pith.science/paper/ARICQWDR"},"agent_actions":{"view_html":"https://pith.science/pith/ARICQWDRUOPEKLEW7L7XMJKPJ5","download_json":"https://pith.science/pith/ARICQWDRUOPEKLEW7L7XMJKPJ5.json","view_paper":"https://pith.science/paper/ARICQWDR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2012.15701&json=true","fetch_graph":"https://pith.science/api/pith-number/ARICQWDRUOPEKLEW7L7XMJKPJ5/graph.json","fetch_events":"https://pith.science/api/pith-number/ARICQWDRUOPEKLEW7L7XMJKPJ5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ARICQWDRUOPEKLEW7L7XMJKPJ5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ARICQWDRUOPEKLEW7L7XMJKPJ5/action/storage_attestation","attest_author":"https://pith.science/pith/ARICQWDRUOPEKLEW7L7XMJKPJ5/action/author_attestation","sign_citation":"https://pith.science/pith/ARICQWDRUOPEKLEW7L7XMJKPJ5/action/citation_signature","submit_replication":"https://pith.science/pith/ARICQWDRUOPEKLEW7L7XMJKPJ5/action/replication_record"}},"created_at":"2026-07-05T02:59:54.975227+00:00","updated_at":"2026-07-05T02:59:54.975227+00:00"}