{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:SM6P4SNZ4M5LH34WEYZPH3QXVT","short_pith_number":"pith:SM6P4SNZ","schema_version":"1.0","canonical_sha256":"933cfe49b9e33ab3ef962632f3ee17acdfe7d5a629ee1144312440009f16c8cf","source":{"kind":"arxiv","id":"2409.03752","version":3},"attestation_state":"computed","paper":{"title":"Attention Heads of Large Language Models: A Survey","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Bo Tang, Feiyu Xiong, Mingchuan Yang, Shichao Song, Yezhaohui Wang, Yuxin Huang, Zhiyu Li, Zifan Zheng","submitted_at":"2024-09-05T17:59:12Z","abstract_excerpt":"Since the advent of ChatGPT, Large Language Models (LLMs) have excelled in various tasks but remain as black-box systems. Understanding the reasoning bottlenecks of LLMs has become a critical challenge, as these limitations are deeply tied to their internal architecture. Among these, attention heads have emerged as a focal point for investigating the underlying mechanics of LLMs. In this survey, we aim to demystify the internal reasoning processes of LLMs by systematically exploring the roles and mechanisms of attention heads. We first introduce a novel four-stage framework inspired by the hum"},"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":"2409.03752","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-09-05T17:59:12Z","cross_cats_sorted":[],"title_canon_sha256":"d0396ab185a1b8f9e71dcd335a2bbe7ace88017a4ca99a27accad31df18ed6d5","abstract_canon_sha256":"5b8771eb15d5e1cb9a25bfedab52e3625cece47f5ede9688e96f898ed4e37355"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:53:04.661793Z","signature_b64":"X5nwvrGS1nKDbL0A/dskQ3XUcGVInT0P9G266gz47pCfE5ZEkBGBAKfSu4cE++sQGhEE7gV5qGM+oqbzNDrqDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"933cfe49b9e33ab3ef962632f3ee17acdfe7d5a629ee1144312440009f16c8cf","last_reissued_at":"2026-07-05T09:53:04.661307Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:53:04.661307Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Attention Heads of Large Language Models: A Survey","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Bo Tang, Feiyu Xiong, Mingchuan Yang, Shichao Song, Yezhaohui Wang, Yuxin Huang, Zhiyu Li, Zifan Zheng","submitted_at":"2024-09-05T17:59:12Z","abstract_excerpt":"Since the advent of ChatGPT, Large Language Models (LLMs) have excelled in various tasks but remain as black-box systems. Understanding the reasoning bottlenecks of LLMs has become a critical challenge, as these limitations are deeply tied to their internal architecture. Among these, attention heads have emerged as a focal point for investigating the underlying mechanics of LLMs. In this survey, we aim to demystify the internal reasoning processes of LLMs by systematically exploring the roles and mechanisms of attention heads. We first introduce a novel four-stage framework inspired by the hum"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.03752","kind":"arxiv","version":3},"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/2409.03752/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":"2409.03752","created_at":"2026-07-05T09:53:04.661373+00:00"},{"alias_kind":"arxiv_version","alias_value":"2409.03752v3","created_at":"2026-07-05T09:53:04.661373+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.03752","created_at":"2026-07-05T09:53:04.661373+00:00"},{"alias_kind":"pith_short_12","alias_value":"SM6P4SNZ4M5L","created_at":"2026-07-05T09:53:04.661373+00:00"},{"alias_kind":"pith_short_16","alias_value":"SM6P4SNZ4M5LH34W","created_at":"2026-07-05T09:53:04.661373+00:00"},{"alias_kind":"pith_short_8","alias_value":"SM6P4SNZ","created_at":"2026-07-05T09:53:04.661373+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":8,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24467","citing_title":"CompressKV: Semantic-Retrieval-Guided KV-Cache Compression for Resource-Efficient Long-Context LLM Inference","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06840","citing_title":"Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2509.24164","citing_title":"Localizing Task Recognition and Task Learning in In-Context Learning via Attention Head Analysis","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2602.11528","citing_title":"Stop Tracking Me! Proactive Defense Against Attribute Inference Attack in LLMs","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03764","citing_title":"Automated Attention Pattern Discovery at Scale in Large Language Models","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10622","citing_title":"Vocabulary Hijacking in LVLMs: Unveiling Critical Attention Heads by Excluding Inert Tokens to Mitigate Hallucination","ref_index":130,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06393","citing_title":"ART: Attention Replacement Technique to Improve Factuality in LLMs","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04641","citing_title":"CAST: Mitigating Object Hallucination in Large Vision-Language Models via Caption-Guided Visual Attention Steering","ref_index":61,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SM6P4SNZ4M5LH34WEYZPH3QXVT","json":"https://pith.science/pith/SM6P4SNZ4M5LH34WEYZPH3QXVT.json","graph_json":"https://pith.science/api/pith-number/SM6P4SNZ4M5LH34WEYZPH3QXVT/graph.json","events_json":"https://pith.science/api/pith-number/SM6P4SNZ4M5LH34WEYZPH3QXVT/events.json","paper":"https://pith.science/paper/SM6P4SNZ"},"agent_actions":{"view_html":"https://pith.science/pith/SM6P4SNZ4M5LH34WEYZPH3QXVT","download_json":"https://pith.science/pith/SM6P4SNZ4M5LH34WEYZPH3QXVT.json","view_paper":"https://pith.science/paper/SM6P4SNZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2409.03752&json=true","fetch_graph":"https://pith.science/api/pith-number/SM6P4SNZ4M5LH34WEYZPH3QXVT/graph.json","fetch_events":"https://pith.science/api/pith-number/SM6P4SNZ4M5LH34WEYZPH3QXVT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SM6P4SNZ4M5LH34WEYZPH3QXVT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SM6P4SNZ4M5LH34WEYZPH3QXVT/action/storage_attestation","attest_author":"https://pith.science/pith/SM6P4SNZ4M5LH34WEYZPH3QXVT/action/author_attestation","sign_citation":"https://pith.science/pith/SM6P4SNZ4M5LH34WEYZPH3QXVT/action/citation_signature","submit_replication":"https://pith.science/pith/SM6P4SNZ4M5LH34WEYZPH3QXVT/action/replication_record"}},"created_at":"2026-07-05T09:53:04.661373+00:00","updated_at":"2026-07-05T09:53:04.661373+00:00"}