{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:FMGYNXEW3PHQWIQKZKX76H5WJG","short_pith_number":"pith:FMGYNXEW","schema_version":"1.0","canonical_sha256":"2b0d86dc96dbcf0b220acaafff1fb6499669d3ebbdbe91cd07f1d4b01ae8d8e6","source":{"kind":"arxiv","id":"2102.12702","version":1},"attestation_state":"computed","paper":{"title":"LazyFormer: Self Attention with Lazy Update","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Chengxuan Ying, Di He, Guolin Ke, Tie-Yan Liu","submitted_at":"2021-02-25T06:18:20Z","abstract_excerpt":"Improving the efficiency of Transformer-based language pre-training is an important task in NLP, especially for the self-attention module, which is computationally expensive. In this paper, we propose a simple but effective solution, called \\emph{LazyFormer}, which computes the self-attention distribution infrequently. LazyFormer composes of multiple lazy blocks, each of which contains multiple Transformer layers. In each lazy block, the self-attention distribution is only computed once in the first layer and then is reused in all upper layers. In this way, the cost of computation could be lar"},"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":"2102.12702","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2021-02-25T06:18:20Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"59574fda5a180a407d1d2bec1037814cd1cb8948e7bfcfb067ace2ce85a8a83f","abstract_canon_sha256":"fe366ecc3f016ef1697a4b44cf5101093b9fab4f42294d2dc99a8ad5d5df9445"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:18:23.053565Z","signature_b64":"byr8C9dipCnRLEpnV0LQyYpk9ciJ7B/FUSlpfmQ/ioX7ERiT1Zm79BCqiJ6kLy160ZBGluQR3yiNcGvUcHQ1Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2b0d86dc96dbcf0b220acaafff1fb6499669d3ebbdbe91cd07f1d4b01ae8d8e6","last_reissued_at":"2026-07-05T02:18:23.053144Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:18:23.053144Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LazyFormer: Self Attention with Lazy Update","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Chengxuan Ying, Di He, Guolin Ke, Tie-Yan Liu","submitted_at":"2021-02-25T06:18:20Z","abstract_excerpt":"Improving the efficiency of Transformer-based language pre-training is an important task in NLP, especially for the self-attention module, which is computationally expensive. In this paper, we propose a simple but effective solution, called \\emph{LazyFormer}, which computes the self-attention distribution infrequently. LazyFormer composes of multiple lazy blocks, each of which contains multiple Transformer layers. In each lazy block, the self-attention distribution is only computed once in the first layer and then is reused in all upper layers. In this way, the cost of computation could be lar"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2102.12702","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/2102.12702/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":"2102.12702","created_at":"2026-07-05T02:18:23.053206+00:00"},{"alias_kind":"arxiv_version","alias_value":"2102.12702v1","created_at":"2026-07-05T02:18:23.053206+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2102.12702","created_at":"2026-07-05T02:18:23.053206+00:00"},{"alias_kind":"pith_short_12","alias_value":"FMGYNXEW3PHQ","created_at":"2026-07-05T02:18:23.053206+00:00"},{"alias_kind":"pith_short_16","alias_value":"FMGYNXEW3PHQWIQK","created_at":"2026-07-05T02:18:23.053206+00:00"},{"alias_kind":"pith_short_8","alias_value":"FMGYNXEW","created_at":"2026-07-05T02:18:23.053206+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2411.12502","citing_title":"Transformer Neural Processes - Kernel Regression","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18529","citing_title":"HybridGen: Efficient LLM Generative Inference via CPU-GPU Hybrid Computing","ref_index":48,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FMGYNXEW3PHQWIQKZKX76H5WJG","json":"https://pith.science/pith/FMGYNXEW3PHQWIQKZKX76H5WJG.json","graph_json":"https://pith.science/api/pith-number/FMGYNXEW3PHQWIQKZKX76H5WJG/graph.json","events_json":"https://pith.science/api/pith-number/FMGYNXEW3PHQWIQKZKX76H5WJG/events.json","paper":"https://pith.science/paper/FMGYNXEW"},"agent_actions":{"view_html":"https://pith.science/pith/FMGYNXEW3PHQWIQKZKX76H5WJG","download_json":"https://pith.science/pith/FMGYNXEW3PHQWIQKZKX76H5WJG.json","view_paper":"https://pith.science/paper/FMGYNXEW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2102.12702&json=true","fetch_graph":"https://pith.science/api/pith-number/FMGYNXEW3PHQWIQKZKX76H5WJG/graph.json","fetch_events":"https://pith.science/api/pith-number/FMGYNXEW3PHQWIQKZKX76H5WJG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FMGYNXEW3PHQWIQKZKX76H5WJG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FMGYNXEW3PHQWIQKZKX76H5WJG/action/storage_attestation","attest_author":"https://pith.science/pith/FMGYNXEW3PHQWIQKZKX76H5WJG/action/author_attestation","sign_citation":"https://pith.science/pith/FMGYNXEW3PHQWIQKZKX76H5WJG/action/citation_signature","submit_replication":"https://pith.science/pith/FMGYNXEW3PHQWIQKZKX76H5WJG/action/replication_record"}},"created_at":"2026-07-05T02:18:23.053206+00:00","updated_at":"2026-07-05T02:18:23.053206+00:00"}