{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:HXHSEMVRTM6S57EBUNLSY222L7","short_pith_number":"pith:HXHSEMVR","schema_version":"1.0","canonical_sha256":"3dcf2232b19b3d2efc81a3572c6b5a5fd1446d2f321232ca24b5f0929c1e78ca","source":{"kind":"arxiv","id":"2403.06764","version":3},"attestation_state":"computed","paper":{"title":"An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CV","authors_text":"Baobao Chang, Chang Zhou, Haozhe Zhao, Junyang Lin, Liang Chen, Shuai Bai, Tianyu Liu","submitted_at":"2024-03-11T14:35:32Z","abstract_excerpt":"In this study, we identify the inefficient attention phenomena in Large Vision-Language Models (LVLMs), notably within prominent models like LLaVA-1.5, QwenVL-Chat and Video-LLaVA. We find out that the attention computation over visual tokens is of extreme inefficiency in the deep layers of popular LVLMs, suggesting a need for a sparser approach compared to textual data handling. To this end, we introduce FastV, a versatile plug-and-play method designed to optimize computational efficiency by learning adaptive attention patterns in early layers and pruning visual tokens in subsequent ones. Our"},"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":"2403.06764","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2024-03-11T14:35:32Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"bd55663b2cc6ca72bfc21c8d8f810908610655392a1be0a7220bd48692b47fac","abstract_canon_sha256":"9d3c9c0425d0d1be97de821c29782d31974c217cd2ad1d65390a6f33f8bd592a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:01:30.929409Z","signature_b64":"HdQQpXMRA4AS2hjhfwhxl4C91CGffM4u8lyZiq+0Zjhddf+ggEa2ZyXW3Z9CgEb9U2fNtF9ZMzZ35RM0G+DYCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3dcf2232b19b3d2efc81a3572c6b5a5fd1446d2f321232ca24b5f0929c1e78ca","last_reissued_at":"2026-07-05T09:01:30.928892Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:01:30.928892Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.CV","authors_text":"Baobao Chang, Chang Zhou, Haozhe Zhao, Junyang Lin, Liang Chen, Shuai Bai, Tianyu Liu","submitted_at":"2024-03-11T14:35:32Z","abstract_excerpt":"In this study, we identify the inefficient attention phenomena in Large Vision-Language Models (LVLMs), notably within prominent models like LLaVA-1.5, QwenVL-Chat and Video-LLaVA. We find out that the attention computation over visual tokens is of extreme inefficiency in the deep layers of popular LVLMs, suggesting a need for a sparser approach compared to textual data handling. To this end, we introduce FastV, a versatile plug-and-play method designed to optimize computational efficiency by learning adaptive attention patterns in early layers and pruning visual tokens in subsequent ones. Our"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.06764","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/2403.06764/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":"2403.06764","created_at":"2026-07-05T09:01:30.928960+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.06764v3","created_at":"2026-07-05T09:01:30.928960+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.06764","created_at":"2026-07-05T09:01:30.928960+00:00"},{"alias_kind":"pith_short_12","alias_value":"HXHSEMVRTM6S","created_at":"2026-07-05T09:01:30.928960+00:00"},{"alias_kind":"pith_short_16","alias_value":"HXHSEMVRTM6S57EB","created_at":"2026-07-05T09:01:30.928960+00:00"},{"alias_kind":"pith_short_8","alias_value":"HXHSEMVR","created_at":"2026-07-05T09:01:30.928960+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2607.08032","citing_title":"What to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and Agents","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2607.05859","citing_title":"AVA-VLM: Adaptive Visual Attention-Vision Language Model for In-the-Wild Construction Site Monitoring","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2605.30904","citing_title":"MergeTok: Unified Continuous and Discrete Visual Tokenization via Token Merging","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09131","citing_title":"Late-Layer Fusion is Enough: Dual-Path Vision Token Routing for Multimodal Large Language Models under Visual Saturation","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2507.21420","citing_title":"ReGATE: Learning Faster and Better with Fewer Tokens in MLLMs","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2410.10781","citing_title":"When Attention Sink Emerges in Language Models: An Empirical View","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2603.07080","citing_title":"VLN-Cache: Enabling Token Caching for VLN Models with Visual/Semantic Dynamics Awareness","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2410.17247","citing_title":"PyramidDrop: Accelerating Your Large Vision-Language Models via Pyramid Visual Redundancy Reduction","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2406.02069","citing_title":"PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12358","citing_title":"Why and When Visual Token Pruning Fails? A Study on Relevant Visual Information Shift in MLLMs Decoding","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09442","citing_title":"UIPress: Bringing Optical Token Compression to UI-to-Code Generation","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HXHSEMVRTM6S57EBUNLSY222L7","json":"https://pith.science/pith/HXHSEMVRTM6S57EBUNLSY222L7.json","graph_json":"https://pith.science/api/pith-number/HXHSEMVRTM6S57EBUNLSY222L7/graph.json","events_json":"https://pith.science/api/pith-number/HXHSEMVRTM6S57EBUNLSY222L7/events.json","paper":"https://pith.science/paper/HXHSEMVR"},"agent_actions":{"view_html":"https://pith.science/pith/HXHSEMVRTM6S57EBUNLSY222L7","download_json":"https://pith.science/pith/HXHSEMVRTM6S57EBUNLSY222L7.json","view_paper":"https://pith.science/paper/HXHSEMVR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.06764&json=true","fetch_graph":"https://pith.science/api/pith-number/HXHSEMVRTM6S57EBUNLSY222L7/graph.json","fetch_events":"https://pith.science/api/pith-number/HXHSEMVRTM6S57EBUNLSY222L7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HXHSEMVRTM6S57EBUNLSY222L7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HXHSEMVRTM6S57EBUNLSY222L7/action/storage_attestation","attest_author":"https://pith.science/pith/HXHSEMVRTM6S57EBUNLSY222L7/action/author_attestation","sign_citation":"https://pith.science/pith/HXHSEMVRTM6S57EBUNLSY222L7/action/citation_signature","submit_replication":"https://pith.science/pith/HXHSEMVRTM6S57EBUNLSY222L7/action/replication_record"}},"created_at":"2026-07-05T09:01:30.928960+00:00","updated_at":"2026-07-05T09:01:30.928960+00:00"}