{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:B7RDMKHK6OR56J3ITYYM2QX5A6","short_pith_number":"pith:B7RDMKHK","schema_version":"1.0","canonical_sha256":"0fe23628eaf3a3df27689e30cd42fd07884f3e15c318d0fe0668e6291a959ecb","source":{"kind":"arxiv","id":"2211.04898","version":2},"attestation_state":"computed","paper":{"title":"Mask More and Mask Later: Efficient Pre-training of Masked Language Models by Disentangling the [MASK] Token","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Baohao Liao, Christof Monz, David Thulke, Hermann Ney, Sanjika Hewavitharana","submitted_at":"2022-11-09T14:03:22Z","abstract_excerpt":"The pre-training of masked language models (MLMs) consumes massive computation to achieve good results on downstream NLP tasks, resulting in a large carbon footprint. In the vanilla MLM, the virtual tokens, [MASK]s, act as placeholders and gather the contextualized information from unmasked tokens to restore the corrupted information. It raises the question of whether we can append [MASK]s at a later layer, to reduce the sequence length for earlier layers and make the pre-training more efficient. We show: (1) [MASK]s can indeed be appended at a later layer, being disentangled from the word emb"},"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":"2211.04898","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-11-09T14:03:22Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"8153be6c094a63a26671379d9f5aa87342555c18a565635ab2fc0098a75b0717","abstract_canon_sha256":"943e4b1a8651546ae19f3fa4641ab417bb3e37b4a0fa672941c0afad5ee0887a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:16:27.351912Z","signature_b64":"lvImafcNns+/d0D6bsfU+YXBMHm+jHLvXUuNpNB1U3J19RuLfzM9vb5wGbh2TDClUjpvOa/WZIvabheVa2etAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0fe23628eaf3a3df27689e30cd42fd07884f3e15c318d0fe0668e6291a959ecb","last_reissued_at":"2026-07-05T05:16:27.351426Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:16:27.351426Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mask More and Mask Later: Efficient Pre-training of Masked Language Models by Disentangling the [MASK] Token","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Baohao Liao, Christof Monz, David Thulke, Hermann Ney, Sanjika Hewavitharana","submitted_at":"2022-11-09T14:03:22Z","abstract_excerpt":"The pre-training of masked language models (MLMs) consumes massive computation to achieve good results on downstream NLP tasks, resulting in a large carbon footprint. In the vanilla MLM, the virtual tokens, [MASK]s, act as placeholders and gather the contextualized information from unmasked tokens to restore the corrupted information. It raises the question of whether we can append [MASK]s at a later layer, to reduce the sequence length for earlier layers and make the pre-training more efficient. We show: (1) [MASK]s can indeed be appended at a later layer, being disentangled from the word emb"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2211.04898","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/2211.04898/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":"2211.04898","created_at":"2026-07-05T05:16:27.351502+00:00"},{"alias_kind":"arxiv_version","alias_value":"2211.04898v2","created_at":"2026-07-05T05:16:27.351502+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2211.04898","created_at":"2026-07-05T05:16:27.351502+00:00"},{"alias_kind":"pith_short_12","alias_value":"B7RDMKHK6OR5","created_at":"2026-07-05T05:16:27.351502+00:00"},{"alias_kind":"pith_short_16","alias_value":"B7RDMKHK6OR56J3I","created_at":"2026-07-05T05:16:27.351502+00:00"},{"alias_kind":"pith_short_8","alias_value":"B7RDMKHK","created_at":"2026-07-05T05:16:27.351502+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2501.13397","citing_title":"ExLM: Rethinking the Impact of [MASK] Tokens in Masked Language Models","ref_index":37,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/B7RDMKHK6OR56J3ITYYM2QX5A6","json":"https://pith.science/pith/B7RDMKHK6OR56J3ITYYM2QX5A6.json","graph_json":"https://pith.science/api/pith-number/B7RDMKHK6OR56J3ITYYM2QX5A6/graph.json","events_json":"https://pith.science/api/pith-number/B7RDMKHK6OR56J3ITYYM2QX5A6/events.json","paper":"https://pith.science/paper/B7RDMKHK"},"agent_actions":{"view_html":"https://pith.science/pith/B7RDMKHK6OR56J3ITYYM2QX5A6","download_json":"https://pith.science/pith/B7RDMKHK6OR56J3ITYYM2QX5A6.json","view_paper":"https://pith.science/paper/B7RDMKHK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2211.04898&json=true","fetch_graph":"https://pith.science/api/pith-number/B7RDMKHK6OR56J3ITYYM2QX5A6/graph.json","fetch_events":"https://pith.science/api/pith-number/B7RDMKHK6OR56J3ITYYM2QX5A6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/B7RDMKHK6OR56J3ITYYM2QX5A6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/B7RDMKHK6OR56J3ITYYM2QX5A6/action/storage_attestation","attest_author":"https://pith.science/pith/B7RDMKHK6OR56J3ITYYM2QX5A6/action/author_attestation","sign_citation":"https://pith.science/pith/B7RDMKHK6OR56J3ITYYM2QX5A6/action/citation_signature","submit_replication":"https://pith.science/pith/B7RDMKHK6OR56J3ITYYM2QX5A6/action/replication_record"}},"created_at":"2026-07-05T05:16:27.351502+00:00","updated_at":"2026-07-05T05:16:27.351502+00:00"}