{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7MRRGVCHBJPYHNTRGYLK5IQSKS","short_pith_number":"pith:7MRRGVCH","schema_version":"1.0","canonical_sha256":"fb231354470a5f83b6713616aea21254b11663975968cbe050b3b75af3e92b7d","source":{"kind":"arxiv","id":"2502.11494","version":2},"attestation_state":"computed","paper":{"title":"Stop Looking for Important Tokens in Multimodal Language Models: Duplication Matters More","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Conghui He, Junyuan Zhang, Linfeng Zhang, Qintong Zhang, Shaobo Wang, Weijia Li, Yifeng Gao, Zichen Wen","submitted_at":"2025-02-17T06:56:28Z","abstract_excerpt":"Vision tokens in multimodal large language models often dominate huge computational overhead due to their excessive length compared to linguistic modality. Abundant recent methods aim to solve this problem with token pruning, which first defines an importance criterion for tokens and then prunes the unimportant vision tokens during inference. However, in this paper, we show that the importance is not an ideal indicator to decide whether a token should be pruned. Surprisingly, it usually results in inferior performance than random token pruning and leading to incompatibility to efficient attent"},"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":"2502.11494","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-02-17T06:56:28Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"e32efbd4b322b7744d5c96b251ef03585c883910019116e583366e95badaf2b5","abstract_canon_sha256":"b5701b55cbce02c5c5b94b2b33c0158c7522a2900db0908443e3a6fee3c3d53e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:17:41.373986Z","signature_b64":"OkTSP8eg/2jixoVChdNKV1bswjY1gyrQWg6MQlg9s/giwR2IrRGamPMli8ki9InAWMQlUhLLlKhee/0GQvTFBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fb231354470a5f83b6713616aea21254b11663975968cbe050b3b75af3e92b7d","last_reissued_at":"2026-07-05T11:17:41.373479Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:17:41.373479Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stop Looking for Important Tokens in Multimodal Language Models: Duplication Matters More","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Conghui He, Junyuan Zhang, Linfeng Zhang, Qintong Zhang, Shaobo Wang, Weijia Li, Yifeng Gao, Zichen Wen","submitted_at":"2025-02-17T06:56:28Z","abstract_excerpt":"Vision tokens in multimodal large language models often dominate huge computational overhead due to their excessive length compared to linguistic modality. Abundant recent methods aim to solve this problem with token pruning, which first defines an importance criterion for tokens and then prunes the unimportant vision tokens during inference. However, in this paper, we show that the importance is not an ideal indicator to decide whether a token should be pruned. Surprisingly, it usually results in inferior performance than random token pruning and leading to incompatibility to efficient attent"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.11494","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/2502.11494/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":"2502.11494","created_at":"2026-07-05T11:17:41.373538+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.11494v2","created_at":"2026-07-05T11:17:41.373538+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.11494","created_at":"2026-07-05T11:17:41.373538+00:00"},{"alias_kind":"pith_short_12","alias_value":"7MRRGVCHBJPY","created_at":"2026-07-05T11:17:41.373538+00:00"},{"alias_kind":"pith_short_16","alias_value":"7MRRGVCHBJPYHNTR","created_at":"2026-07-05T11:17:41.373538+00:00"},{"alias_kind":"pith_short_8","alias_value":"7MRRGVCH","created_at":"2026-07-05T11:17:41.373538+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":14,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.27660","citing_title":"MVPruner: Dynamic Token Pruning for Accelerating Multi-view Vision-Language Models in Autonomous Driving","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03569","citing_title":"When Attention Collapses: Stage-Aware Visual Token Pruning from Structure to Semantics","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31383","citing_title":"MS-Resampler: Multi-Scope Visual Resampling for Efficient Multimodal LLMs","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2606.27660","citing_title":"MVPruner: Dynamic Token Pruning for Accelerating Multi-view Vision-Language Models in Autonomous Driving","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00392","citing_title":"RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20950","citing_title":"Focus-then-Context: Subject-Centric Progressive Visual Token Reduction for Vision-Language Models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2603.01400","citing_title":"Token Reduction via Local and Global Contexts Optimization for Efficient Video Large Language Models","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09649","citing_title":"Make Each Token Count: Towards Improving Long-Context Performance with KV Cache Eviction","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00392","citing_title":"RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23950","citing_title":"LearnPruner: Rethinking Attention-based Token Pruning in Vision Language Models","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00392","citing_title":"RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19145","citing_title":"ST-Prune: Training-Free Spatio-Temporal Token Pruning for Vision-Language Models in Autonomous Driving","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07812","citing_title":"HAWK: Head Importance-Aware Visual Token Pruning in Multimodal Models","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17087","citing_title":"EvoComp: Learning Visual Token Compression for Multimodal Large Language Models via Semantic-Guided Evolutionary Labeling","ref_index":51,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7MRRGVCHBJPYHNTRGYLK5IQSKS","json":"https://pith.science/pith/7MRRGVCHBJPYHNTRGYLK5IQSKS.json","graph_json":"https://pith.science/api/pith-number/7MRRGVCHBJPYHNTRGYLK5IQSKS/graph.json","events_json":"https://pith.science/api/pith-number/7MRRGVCHBJPYHNTRGYLK5IQSKS/events.json","paper":"https://pith.science/paper/7MRRGVCH"},"agent_actions":{"view_html":"https://pith.science/pith/7MRRGVCHBJPYHNTRGYLK5IQSKS","download_json":"https://pith.science/pith/7MRRGVCHBJPYHNTRGYLK5IQSKS.json","view_paper":"https://pith.science/paper/7MRRGVCH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.11494&json=true","fetch_graph":"https://pith.science/api/pith-number/7MRRGVCHBJPYHNTRGYLK5IQSKS/graph.json","fetch_events":"https://pith.science/api/pith-number/7MRRGVCHBJPYHNTRGYLK5IQSKS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7MRRGVCHBJPYHNTRGYLK5IQSKS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7MRRGVCHBJPYHNTRGYLK5IQSKS/action/storage_attestation","attest_author":"https://pith.science/pith/7MRRGVCHBJPYHNTRGYLK5IQSKS/action/author_attestation","sign_citation":"https://pith.science/pith/7MRRGVCHBJPYHNTRGYLK5IQSKS/action/citation_signature","submit_replication":"https://pith.science/pith/7MRRGVCHBJPYHNTRGYLK5IQSKS/action/replication_record"}},"created_at":"2026-07-05T11:17:41.373538+00:00","updated_at":"2026-07-05T11:17:41.373538+00:00"}