{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:DKZEWOV6X2FQGQSK623D23SCKN","short_pith_number":"pith:DKZEWOV6","schema_version":"1.0","canonical_sha256":"1ab24b3abebe8b03424af6b63d6e425363f772789e5685b9b395b752c0f475c2","source":{"kind":"arxiv","id":"2105.11618","version":1},"attestation_state":"computed","paper":{"title":"TR-BERT: Dynamic Token Reduction for Accelerating BERT Inference","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Deming Ye, Maosong Sun, Yankai Lin, Yufei Huang","submitted_at":"2021-05-25T02:28:51Z","abstract_excerpt":"Existing pre-trained language models (PLMs) are often computationally expensive in inference, making them impractical in various resource-limited real-world applications. To address this issue, we propose a dynamic token reduction approach to accelerate PLMs' inference, named TR-BERT, which could flexibly adapt the layer number of each token in inference to avoid redundant calculation. Specially, TR-BERT formulates the token reduction process as a multi-step token selection problem and automatically learns the selection strategy via reinforcement learning. The experimental results on several d"},"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":"2105.11618","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-05-25T02:28:51Z","cross_cats_sorted":[],"title_canon_sha256":"679b78de5c7b49cf7adb29894dcbf49d27b50550015b5e53343c00905e9acdde","abstract_canon_sha256":"ad7f782fdc8ef4ebe35bc673c1156bcf813cd491c6197952f24c0cb25938e476"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:42:54.491250Z","signature_b64":"rzRyjk+SfAv9QqMAwMMSxldweThGRtu5Ij3OIq0C96ipvjM85ts5K8/YOnuMg3/Wwg8DRCE09Pn4Io6fut0VCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1ab24b3abebe8b03424af6b63d6e425363f772789e5685b9b395b752c0f475c2","last_reissued_at":"2026-07-05T02:42:54.490798Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:42:54.490798Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TR-BERT: Dynamic Token Reduction for Accelerating BERT Inference","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Deming Ye, Maosong Sun, Yankai Lin, Yufei Huang","submitted_at":"2021-05-25T02:28:51Z","abstract_excerpt":"Existing pre-trained language models (PLMs) are often computationally expensive in inference, making them impractical in various resource-limited real-world applications. To address this issue, we propose a dynamic token reduction approach to accelerate PLMs' inference, named TR-BERT, which could flexibly adapt the layer number of each token in inference to avoid redundant calculation. Specially, TR-BERT formulates the token reduction process as a multi-step token selection problem and automatically learns the selection strategy via reinforcement learning. The experimental results on several d"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2105.11618","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/2105.11618/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":"2105.11618","created_at":"2026-07-05T02:42:54.490863+00:00"},{"alias_kind":"arxiv_version","alias_value":"2105.11618v1","created_at":"2026-07-05T02:42:54.490863+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2105.11618","created_at":"2026-07-05T02:42:54.490863+00:00"},{"alias_kind":"pith_short_12","alias_value":"DKZEWOV6X2FQ","created_at":"2026-07-05T02:42:54.490863+00:00"},{"alias_kind":"pith_short_16","alias_value":"DKZEWOV6X2FQGQSK","created_at":"2026-07-05T02:42:54.490863+00:00"},{"alias_kind":"pith_short_8","alias_value":"DKZEWOV6","created_at":"2026-07-05T02:42:54.490863+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2508.15190","citing_title":"SemToken: Semantic-Aware Tokenization for Efficient Long-Context Language Modeling","ref_index":35,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DKZEWOV6X2FQGQSK623D23SCKN","json":"https://pith.science/pith/DKZEWOV6X2FQGQSK623D23SCKN.json","graph_json":"https://pith.science/api/pith-number/DKZEWOV6X2FQGQSK623D23SCKN/graph.json","events_json":"https://pith.science/api/pith-number/DKZEWOV6X2FQGQSK623D23SCKN/events.json","paper":"https://pith.science/paper/DKZEWOV6"},"agent_actions":{"view_html":"https://pith.science/pith/DKZEWOV6X2FQGQSK623D23SCKN","download_json":"https://pith.science/pith/DKZEWOV6X2FQGQSK623D23SCKN.json","view_paper":"https://pith.science/paper/DKZEWOV6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2105.11618&json=true","fetch_graph":"https://pith.science/api/pith-number/DKZEWOV6X2FQGQSK623D23SCKN/graph.json","fetch_events":"https://pith.science/api/pith-number/DKZEWOV6X2FQGQSK623D23SCKN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DKZEWOV6X2FQGQSK623D23SCKN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DKZEWOV6X2FQGQSK623D23SCKN/action/storage_attestation","attest_author":"https://pith.science/pith/DKZEWOV6X2FQGQSK623D23SCKN/action/author_attestation","sign_citation":"https://pith.science/pith/DKZEWOV6X2FQGQSK623D23SCKN/action/citation_signature","submit_replication":"https://pith.science/pith/DKZEWOV6X2FQGQSK623D23SCKN/action/replication_record"}},"created_at":"2026-07-05T02:42:54.490863+00:00","updated_at":"2026-07-05T02:42:54.490863+00:00"}