{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:YKIXVOEWAEQMUEYKAR3CWVFJWE","short_pith_number":"pith:YKIXVOEW","schema_version":"1.0","canonical_sha256":"c2917ab8960120ca130a04762b54a9b12e9716f7ebcf5c75087930514d8ae5e7","source":{"kind":"arxiv","id":"2412.20105","version":1},"attestation_state":"computed","paper":{"title":"ST$^3$: Accelerating Multimodal Large Language Model by Spatial-Temporal Visual Token Trimming","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Haoji Hu, Jian Chen, Jiedong Zhuang, Lu Lu, Ming Dai, Qiang Liu, Rui Hu","submitted_at":"2024-12-28T10:17:29Z","abstract_excerpt":"Multimodal large language models (MLLMs) enhance their perceptual capabilities by integrating visual and textual information. However, processing the massive number of visual tokens incurs a significant computational cost. Existing analysis of the MLLM attention mechanisms remains shallow, leading to coarse-grain token pruning strategies that fail to effectively balance speed and accuracy. In this paper, we conduct a comprehensive investigation of MLLM attention mechanisms with LLaVA. We find that numerous visual tokens and partial attention computations are redundant during the decoding proce"},"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":"2412.20105","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2024-12-28T10:17:29Z","cross_cats_sorted":[],"title_canon_sha256":"d5570c9b1a4da5f6179835ec3b4279f01cbb60944c5b7b96336fa453527a33d3","abstract_canon_sha256":"5605af4ca86aea70a0f715ce537fb5ec5ca4511fc9d4e20d55628989ba632bfc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:54:54.711147Z","signature_b64":"Cqw1eO2Gug75ZJtCmsQlwnq/7mTXjmbnbO2xfIgZdvZvBDDQgVK1YOEuiHfvGkQ26pK1ezBtDO1EHcPTO3MZCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c2917ab8960120ca130a04762b54a9b12e9716f7ebcf5c75087930514d8ae5e7","last_reissued_at":"2026-07-05T09:54:54.710671Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:54:54.710671Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ST$^3$: Accelerating Multimodal Large Language Model by Spatial-Temporal Visual Token Trimming","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Haoji Hu, Jian Chen, Jiedong Zhuang, Lu Lu, Ming Dai, Qiang Liu, Rui Hu","submitted_at":"2024-12-28T10:17:29Z","abstract_excerpt":"Multimodal large language models (MLLMs) enhance their perceptual capabilities by integrating visual and textual information. However, processing the massive number of visual tokens incurs a significant computational cost. Existing analysis of the MLLM attention mechanisms remains shallow, leading to coarse-grain token pruning strategies that fail to effectively balance speed and accuracy. In this paper, we conduct a comprehensive investigation of MLLM attention mechanisms with LLaVA. We find that numerous visual tokens and partial attention computations are redundant during the decoding proce"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.20105","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/2412.20105/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":"2412.20105","created_at":"2026-07-05T09:54:54.710728+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.20105v1","created_at":"2026-07-05T09:54:54.710728+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.20105","created_at":"2026-07-05T09:54:54.710728+00:00"},{"alias_kind":"pith_short_12","alias_value":"YKIXVOEWAEQM","created_at":"2026-07-05T09:54:54.710728+00:00"},{"alias_kind":"pith_short_16","alias_value":"YKIXVOEWAEQMUEYK","created_at":"2026-07-05T09:54:54.710728+00:00"},{"alias_kind":"pith_short_8","alias_value":"YKIXVOEW","created_at":"2026-07-05T09:54:54.710728+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21893","citing_title":"AugCodec: A Low-Bitrate Disentangled Neural Speech Codec via Data Augmentation","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YKIXVOEWAEQMUEYKAR3CWVFJWE","json":"https://pith.science/pith/YKIXVOEWAEQMUEYKAR3CWVFJWE.json","graph_json":"https://pith.science/api/pith-number/YKIXVOEWAEQMUEYKAR3CWVFJWE/graph.json","events_json":"https://pith.science/api/pith-number/YKIXVOEWAEQMUEYKAR3CWVFJWE/events.json","paper":"https://pith.science/paper/YKIXVOEW"},"agent_actions":{"view_html":"https://pith.science/pith/YKIXVOEWAEQMUEYKAR3CWVFJWE","download_json":"https://pith.science/pith/YKIXVOEWAEQMUEYKAR3CWVFJWE.json","view_paper":"https://pith.science/paper/YKIXVOEW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.20105&json=true","fetch_graph":"https://pith.science/api/pith-number/YKIXVOEWAEQMUEYKAR3CWVFJWE/graph.json","fetch_events":"https://pith.science/api/pith-number/YKIXVOEWAEQMUEYKAR3CWVFJWE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YKIXVOEWAEQMUEYKAR3CWVFJWE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YKIXVOEWAEQMUEYKAR3CWVFJWE/action/storage_attestation","attest_author":"https://pith.science/pith/YKIXVOEWAEQMUEYKAR3CWVFJWE/action/author_attestation","sign_citation":"https://pith.science/pith/YKIXVOEWAEQMUEYKAR3CWVFJWE/action/citation_signature","submit_replication":"https://pith.science/pith/YKIXVOEWAEQMUEYKAR3CWVFJWE/action/replication_record"}},"created_at":"2026-07-05T09:54:54.710728+00:00","updated_at":"2026-07-05T09:54:54.710728+00:00"}