{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:NCY5SNFAYFKUU5EJI7G4EF7DLA","short_pith_number":"pith:NCY5SNFA","schema_version":"1.0","canonical_sha256":"68b1d934a0c1554a748947cdc217e3581a73640063539835b9e47f2eea4472a4","source":{"kind":"arxiv","id":"1909.10351","version":5},"attestation_state":"computed","paper":{"title":"TinyBERT: Distilling BERT for Natural Language Understanding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Fang Wang, Lifeng Shang, Linlin Li, Qun Liu, Xiao Chen, Xiaoqi Jiao, Xin Jiang, Yichun Yin","submitted_at":"2019-09-23T13:05:35Z","abstract_excerpt":"Language model pre-training, such as BERT, has significantly improved the performances of many natural language processing tasks. However, pre-trained language models are usually computationally expensive, so it is difficult to efficiently execute them on resource-restricted devices. To accelerate inference and reduce model size while maintaining accuracy, we first propose a novel Transformer distillation method that is specially designed for knowledge distillation (KD) of the Transformer-based models. By leveraging this new KD method, the plenty of knowledge encoded in a large teacher BERT ca"},"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":"1909.10351","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2019-09-23T13:05:35Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"a93650afa01faf2bfaedf6780f376ad228f94573026fb6e5fbb9c4d1f1af6a81","abstract_canon_sha256":"c88719573d2870577ed0a9e6bbd1e81c391c04387f16885b1a1b419e987485ad"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T01:43:23.502409Z","signature_b64":"npANlXlKm3hq6slLVU7gz+p5lBcMG2m7yqrCyaVuQbmVbs2fFyMvQxXdx14hwUsvB8Fdy8Pv9Z2LYAjsTqSEBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"68b1d934a0c1554a748947cdc217e3581a73640063539835b9e47f2eea4472a4","last_reissued_at":"2026-07-05T01:43:23.501588Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T01:43:23.501588Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TinyBERT: Distilling BERT for Natural Language Understanding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Fang Wang, Lifeng Shang, Linlin Li, Qun Liu, Xiao Chen, Xiaoqi Jiao, Xin Jiang, Yichun Yin","submitted_at":"2019-09-23T13:05:35Z","abstract_excerpt":"Language model pre-training, such as BERT, has significantly improved the performances of many natural language processing tasks. However, pre-trained language models are usually computationally expensive, so it is difficult to efficiently execute them on resource-restricted devices. To accelerate inference and reduce model size while maintaining accuracy, we first propose a novel Transformer distillation method that is specially designed for knowledge distillation (KD) of the Transformer-based models. By leveraging this new KD method, the plenty of knowledge encoded in a large teacher BERT ca"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1909.10351","kind":"arxiv","version":5},"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/1909.10351/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":"1909.10351","created_at":"2026-07-05T01:43:23.501649+00:00"},{"alias_kind":"arxiv_version","alias_value":"1909.10351v5","created_at":"2026-07-05T01:43:23.501649+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1909.10351","created_at":"2026-07-05T01:43:23.501649+00:00"},{"alias_kind":"pith_short_12","alias_value":"NCY5SNFAYFKU","created_at":"2026-07-05T01:43:23.501649+00:00"},{"alias_kind":"pith_short_16","alias_value":"NCY5SNFAYFKUU5EJ","created_at":"2026-07-05T01:43:23.501649+00:00"},{"alias_kind":"pith_short_8","alias_value":"NCY5SNFA","created_at":"2026-07-05T01:43:23.501649+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":29,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23919","citing_title":"Unified Multi-Task Relevance Modeling for E-Commerce: Comparing Task Routing Architectures Across LLMs and Cross-Encoders","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21851","citing_title":"TALAS: Teacher-Anchored Layer Alignment with Adaptive Sharpness-Aware Minimization for Embedding Distillation","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09401","citing_title":"Benchmarking Empirical Privacy Protection for Adaptations of Large Language Models","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00382","citing_title":"Vitality-Aware Compression for Efficient Image-to-Shape Diffusion Transformers","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05429","citing_title":"Minimizing the Hidden Cost of Scales: Graph-Guided Ultra-Low-Bit Quantization for Large Language Models","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04032","citing_title":"Do Transformers Need Three Projections? Systematic Study of QKV Variants","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01374","citing_title":"MTA: Multi-Granular Trajectory Alignment for Large Language Model Distillation","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24667","citing_title":"When Mean CE Fails: Median CE Can Better Track Language Model Quality","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28386","citing_title":"Data Provenance for Image Auto-Regressive Generation","ref_index":106,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30448","citing_title":"Bounded Behavioral Indistinguishability for Black-Box LLM Distillation","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09881","citing_title":"Toward Calibrated, Fair, and accurate Deepfake Detection","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2503.08223","citing_title":"Will LLMs Scaling Hit the Wall? Breaking Barriers via Distributed Resources on Massive Edge Devices","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19842","citing_title":"Fast Tensorization of Neural Networks via Slice-wise Feature Distillation","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2508.03949","citing_title":"Model Compression vs. Adversarial Robustness: An Empirical Study on Language Models for Code","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2509.11443","citing_title":"A Transformer-Based Cross-Platform Analysis of Public Discourse on the 15-Minute City Paradigm","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2511.05476","citing_title":"A Metamorphic Testing Perspective on Knowledge Distillation for Language Models of Code: Does the Student Deeply Mimic the Teacher?","ref_index":79,"is_internal_anchor":false},{"citing_arxiv_id":"2101.02388","citing_title":"Knowledge Distillation in Iterative Generative Models for Improved Sampling Speed","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2003.10555","citing_title":"ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2111.09543","citing_title":"DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient-Disentangled Embedding Sharing","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13859","citing_title":"BiSpikCLM: A Spiking Language Model integrating Softmax-Free Spiking Attention and Spike-Aware Alignment Distillation","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13421","citing_title":"Combining pre-trained models via localized model averaging","ref_index":150,"is_internal_anchor":false},{"citing_arxiv_id":"1910.10683","citing_title":"Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01374","citing_title":"MTA: Multi-Granular Trajectory Alignment for Large Language Model Distillation","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01205","citing_title":"SRA: Span Representation Alignment for Large Language Model Distillation","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18939","citing_title":"TabEmb: Joint Semantic-Structure Embedding for Table Annotation","ref_index":42,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NCY5SNFAYFKUU5EJI7G4EF7DLA","json":"https://pith.science/pith/NCY5SNFAYFKUU5EJI7G4EF7DLA.json","graph_json":"https://pith.science/api/pith-number/NCY5SNFAYFKUU5EJI7G4EF7DLA/graph.json","events_json":"https://pith.science/api/pith-number/NCY5SNFAYFKUU5EJI7G4EF7DLA/events.json","paper":"https://pith.science/paper/NCY5SNFA"},"agent_actions":{"view_html":"https://pith.science/pith/NCY5SNFAYFKUU5EJI7G4EF7DLA","download_json":"https://pith.science/pith/NCY5SNFAYFKUU5EJI7G4EF7DLA.json","view_paper":"https://pith.science/paper/NCY5SNFA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1909.10351&json=true","fetch_graph":"https://pith.science/api/pith-number/NCY5SNFAYFKUU5EJI7G4EF7DLA/graph.json","fetch_events":"https://pith.science/api/pith-number/NCY5SNFAYFKUU5EJI7G4EF7DLA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NCY5SNFAYFKUU5EJI7G4EF7DLA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NCY5SNFAYFKUU5EJI7G4EF7DLA/action/storage_attestation","attest_author":"https://pith.science/pith/NCY5SNFAYFKUU5EJI7G4EF7DLA/action/author_attestation","sign_citation":"https://pith.science/pith/NCY5SNFAYFKUU5EJI7G4EF7DLA/action/citation_signature","submit_replication":"https://pith.science/pith/NCY5SNFAYFKUU5EJI7G4EF7DLA/action/replication_record"}},"created_at":"2026-07-05T01:43:23.501649+00:00","updated_at":"2026-07-05T01:43:23.501649+00:00"}