{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:VXVP7CEJPMV3XF3L6NIMGEC44V","short_pith_number":"pith:VXVP7CEJ","schema_version":"1.0","canonical_sha256":"adeaff88897b2bbb976bf350c3105ce5437e3aa4aa314795cd6c3e8eb0ef5eb2","source":{"kind":"arxiv","id":"2012.15828","version":2},"attestation_state":"computed","paper":{"title":"MiniLMv2: Multi-Head Self-Attention Relation Distillation for Compressing Pretrained Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Furu Wei, Hangbo Bao, Li Dong, Shaohan Huang, Wenhui Wang","submitted_at":"2020-12-31T18:51:26Z","abstract_excerpt":"We generalize deep self-attention distillation in MiniLM (Wang et al., 2020) by only using self-attention relation distillation for task-agnostic compression of pretrained Transformers. In particular, we define multi-head self-attention relations as scaled dot-product between the pairs of query, key, and value vectors within each self-attention module. Then we employ the above relational knowledge to train the student model. Besides its simplicity and unified principle, more favorably, there is no restriction in terms of the number of student's attention heads, while most previous work has to "},"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":"2012.15828","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-12-31T18:51:26Z","cross_cats_sorted":[],"title_canon_sha256":"e9b4f72ca219b90dfa5c051fc895c47c2582bf5ade695a72358a75f3a3228e8b","abstract_canon_sha256":"d71dcbcb909ed998e719e0e09a79c64a60454f920e5d4174830df45695f5bd96"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:52:41.494732Z","signature_b64":"4VyAOjJA4gmBjzQ8tM5QWOIsLtegRM9kGbsOUBjId6fN9TdR29ohbsRZfBl/7UuEuMK8vRz/J+01p93P4djQAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"adeaff88897b2bbb976bf350c3105ce5437e3aa4aa314795cd6c3e8eb0ef5eb2","last_reissued_at":"2026-07-05T02:52:41.494206Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:52:41.494206Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MiniLMv2: Multi-Head Self-Attention Relation Distillation for Compressing Pretrained Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Furu Wei, Hangbo Bao, Li Dong, Shaohan Huang, Wenhui Wang","submitted_at":"2020-12-31T18:51:26Z","abstract_excerpt":"We generalize deep self-attention distillation in MiniLM (Wang et al., 2020) by only using self-attention relation distillation for task-agnostic compression of pretrained Transformers. In particular, we define multi-head self-attention relations as scaled dot-product between the pairs of query, key, and value vectors within each self-attention module. Then we employ the above relational knowledge to train the student model. Besides its simplicity and unified principle, more favorably, there is no restriction in terms of the number of student's attention heads, while most previous work has to "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2012.15828","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/2012.15828/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":"2012.15828","created_at":"2026-07-05T02:52:41.494267+00:00"},{"alias_kind":"arxiv_version","alias_value":"2012.15828v2","created_at":"2026-07-05T02:52:41.494267+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2012.15828","created_at":"2026-07-05T02:52:41.494267+00:00"},{"alias_kind":"pith_short_12","alias_value":"VXVP7CEJPMV3","created_at":"2026-07-05T02:52:41.494267+00:00"},{"alias_kind":"pith_short_16","alias_value":"VXVP7CEJPMV3XF3L","created_at":"2026-07-05T02:52:41.494267+00:00"},{"alias_kind":"pith_short_8","alias_value":"VXVP7CEJ","created_at":"2026-07-05T02:52:41.494267+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.29522","citing_title":"DeepSurvey: Enhancing Analytical Depth and Citation Reliability in Automated Survey Generation","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07604","citing_title":"Contribution Weights: A Geometrical Analysis of Self-Attention Transformers","ref_index":103,"is_internal_anchor":false},{"citing_arxiv_id":"2510.17934","citing_title":"AtlasKV: Augmenting LLMs with Billion-Scale Knowledge Graphs in 20GB VRAM","ref_index":33,"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":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24380","citing_title":"Structural Pruning of Large Vision Language Models: A Comprehensive Study on Pruning Dynamics, Recovery, and Data Efficiency","ref_index":41,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VXVP7CEJPMV3XF3L6NIMGEC44V","json":"https://pith.science/pith/VXVP7CEJPMV3XF3L6NIMGEC44V.json","graph_json":"https://pith.science/api/pith-number/VXVP7CEJPMV3XF3L6NIMGEC44V/graph.json","events_json":"https://pith.science/api/pith-number/VXVP7CEJPMV3XF3L6NIMGEC44V/events.json","paper":"https://pith.science/paper/VXVP7CEJ"},"agent_actions":{"view_html":"https://pith.science/pith/VXVP7CEJPMV3XF3L6NIMGEC44V","download_json":"https://pith.science/pith/VXVP7CEJPMV3XF3L6NIMGEC44V.json","view_paper":"https://pith.science/paper/VXVP7CEJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2012.15828&json=true","fetch_graph":"https://pith.science/api/pith-number/VXVP7CEJPMV3XF3L6NIMGEC44V/graph.json","fetch_events":"https://pith.science/api/pith-number/VXVP7CEJPMV3XF3L6NIMGEC44V/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VXVP7CEJPMV3XF3L6NIMGEC44V/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VXVP7CEJPMV3XF3L6NIMGEC44V/action/storage_attestation","attest_author":"https://pith.science/pith/VXVP7CEJPMV3XF3L6NIMGEC44V/action/author_attestation","sign_citation":"https://pith.science/pith/VXVP7CEJPMV3XF3L6NIMGEC44V/action/citation_signature","submit_replication":"https://pith.science/pith/VXVP7CEJPMV3XF3L6NIMGEC44V/action/replication_record"}},"created_at":"2026-07-05T02:52:41.494267+00:00","updated_at":"2026-07-05T02:52:41.494267+00:00"}