{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:RJ6NOCH4L7PMKRJ7C2SD5HWBWZ","short_pith_number":"pith:RJ6NOCH4","schema_version":"1.0","canonical_sha256":"8a7cd708fc5fdec5453f16a43e9ec1b6527b7086da6140757dec5f4db69afa39","source":{"kind":"arxiv","id":"2101.09755","version":1},"attestation_state":"computed","paper":{"title":"RomeBERT: Robust Training of Multi-Exit BERT","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Peng Gao, Shijie Geng, Yongfeng Zhang, Zuohui Fu","submitted_at":"2021-01-24T17:03:57Z","abstract_excerpt":"BERT has achieved superior performances on Natural Language Understanding (NLU) tasks. However, BERT possesses a large number of parameters and demands certain resources to deploy. For acceleration, Dynamic Early Exiting for BERT (DeeBERT) has been proposed recently, which incorporates multiple exits and adopts a dynamic early-exit mechanism to ensure efficient inference. While obtaining an efficiency-performance tradeoff, the performances of early exits in multi-exit BERT are significantly worse than late exits. In this paper, we leverage gradient regularized self-distillation for RObust trai"},"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":"2101.09755","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-01-24T17:03:57Z","cross_cats_sorted":[],"title_canon_sha256":"f8caf4bbf98e1f8546feafff28a96506d9bc1e4fffbfd5774220fc35879f353c","abstract_canon_sha256":"dad15e6320196d23997e057f5beeb9e0bd2a2346f300febabe2b278fa6172fc1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:09:22.218538Z","signature_b64":"EmfHAaYD9FJVyJ0JRugOh8jOWElkqqNM3KY4sMEd4WIhcW06lfKEDHRQUA8kSDH2iv5WJIBoX0wab+0hIbNmDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8a7cd708fc5fdec5453f16a43e9ec1b6527b7086da6140757dec5f4db69afa39","last_reissued_at":"2026-07-05T02:09:22.218150Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:09:22.218150Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RomeBERT: Robust Training of Multi-Exit BERT","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Peng Gao, Shijie Geng, Yongfeng Zhang, Zuohui Fu","submitted_at":"2021-01-24T17:03:57Z","abstract_excerpt":"BERT has achieved superior performances on Natural Language Understanding (NLU) tasks. However, BERT possesses a large number of parameters and demands certain resources to deploy. For acceleration, Dynamic Early Exiting for BERT (DeeBERT) has been proposed recently, which incorporates multiple exits and adopts a dynamic early-exit mechanism to ensure efficient inference. While obtaining an efficiency-performance tradeoff, the performances of early exits in multi-exit BERT are significantly worse than late exits. In this paper, we leverage gradient regularized self-distillation for RObust trai"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2101.09755","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/2101.09755/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":"2101.09755","created_at":"2026-07-05T02:09:22.218214+00:00"},{"alias_kind":"arxiv_version","alias_value":"2101.09755v1","created_at":"2026-07-05T02:09:22.218214+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2101.09755","created_at":"2026-07-05T02:09:22.218214+00:00"},{"alias_kind":"pith_short_12","alias_value":"RJ6NOCH4L7PM","created_at":"2026-07-05T02:09:22.218214+00:00"},{"alias_kind":"pith_short_16","alias_value":"RJ6NOCH4L7PMKRJ7","created_at":"2026-07-05T02:09:22.218214+00:00"},{"alias_kind":"pith_short_8","alias_value":"RJ6NOCH4","created_at":"2026-07-05T02:09:22.218214+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2502.00745","citing_title":"BEEM: Boosting Performance of Early Exit DNNs using Multi-Exit Classifiers as Experts","ref_index":9,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ","json":"https://pith.science/pith/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ.json","graph_json":"https://pith.science/api/pith-number/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ/graph.json","events_json":"https://pith.science/api/pith-number/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ/events.json","paper":"https://pith.science/paper/RJ6NOCH4"},"agent_actions":{"view_html":"https://pith.science/pith/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ","download_json":"https://pith.science/pith/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ.json","view_paper":"https://pith.science/paper/RJ6NOCH4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2101.09755&json=true","fetch_graph":"https://pith.science/api/pith-number/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ/graph.json","fetch_events":"https://pith.science/api/pith-number/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ/action/storage_attestation","attest_author":"https://pith.science/pith/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ/action/author_attestation","sign_citation":"https://pith.science/pith/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ/action/citation_signature","submit_replication":"https://pith.science/pith/RJ6NOCH4L7PMKRJ7C2SD5HWBWZ/action/replication_record"}},"created_at":"2026-07-05T02:09:22.218214+00:00","updated_at":"2026-07-05T02:09:22.218214+00:00"}