{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SZO3DK57JMPFAFFQXYMNT3IPPT","short_pith_number":"pith:SZO3DK57","schema_version":"1.0","canonical_sha256":"965db1abbf4b1e5014b0be18d9ed0f7cc2a66ec38481af777e3263742271626c","source":{"kind":"arxiv","id":"2503.11549","version":1},"attestation_state":"computed","paper":{"title":"Similarity-Aware Token Pruning: Your VLM but Faster","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Ahmadreza Jeddi, Babak Taati, Elham Dolatabadi, Negin Baghbanzadeh","submitted_at":"2025-03-14T16:12:23Z","abstract_excerpt":"The computational demands of Vision Transformers (ViTs) and Vision-Language Models (VLMs) remain a significant challenge due to the quadratic complexity of self-attention. While token pruning offers a promising solution, existing methods often introduce training overhead or fail to adapt dynamically across layers. We present SAINT, a training-free token pruning framework that leverages token similarity and a graph-based formulation to dynamically optimize pruning rates and redundancy thresholds. Through systematic analysis, we identify a universal three-stage token evolution process (aligner-e"},"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":"2503.11549","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-03-14T16:12:23Z","cross_cats_sorted":[],"title_canon_sha256":"80baee3ebe67a8ba25817a78dd7079cbd5ad1807db8e9b4c4d00fafdc1e8e2af","abstract_canon_sha256":"7579c45e0907ad90aaa66ff9b81a6c11b7b344fc9acef2c94c9e7ce353b56d56"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:31:34.301094Z","signature_b64":"qI23C2OiblxRSjWMJxwkOIo1emtz++CpBLOSGchFmm6lhAqiHBh4hriqP69F9yAuyTXn8Y/RvkLxpeiRk4nXDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"965db1abbf4b1e5014b0be18d9ed0f7cc2a66ec38481af777e3263742271626c","last_reissued_at":"2026-07-05T10:31:34.300564Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:31:34.300564Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Similarity-Aware Token Pruning: Your VLM but Faster","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Ahmadreza Jeddi, Babak Taati, Elham Dolatabadi, Negin Baghbanzadeh","submitted_at":"2025-03-14T16:12:23Z","abstract_excerpt":"The computational demands of Vision Transformers (ViTs) and Vision-Language Models (VLMs) remain a significant challenge due to the quadratic complexity of self-attention. While token pruning offers a promising solution, existing methods often introduce training overhead or fail to adapt dynamically across layers. We present SAINT, a training-free token pruning framework that leverages token similarity and a graph-based formulation to dynamically optimize pruning rates and redundancy thresholds. Through systematic analysis, we identify a universal three-stage token evolution process (aligner-e"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.11549","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/2503.11549/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":"2503.11549","created_at":"2026-07-05T10:31:34.300630+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.11549v1","created_at":"2026-07-05T10:31:34.300630+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.11549","created_at":"2026-07-05T10:31:34.300630+00:00"},{"alias_kind":"pith_short_12","alias_value":"SZO3DK57JMPF","created_at":"2026-07-05T10:31:34.300630+00:00"},{"alias_kind":"pith_short_16","alias_value":"SZO3DK57JMPFAFFQ","created_at":"2026-07-05T10:31:34.300630+00:00"},{"alias_kind":"pith_short_8","alias_value":"SZO3DK57","created_at":"2026-07-05T10:31:34.300630+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24165","citing_title":"Spectral Evolution-Guided Token Pruning in Multimodal Large Language Models","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2606.24156","citing_title":"Accelerating Multimodal Large Language Models with Prior-Corrected Token Reduction","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2606.27161","citing_title":"TOPS: First-Principles Visual Token Pruning via Constructing Token Optimal Preservation Sets for Efficient MLLM Inference","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11576","citing_title":"AVIS: Adaptive Test-Time Scaling for Vision-Language Models","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08156","citing_title":"RAPID: Layer-Wise Redundancy-Aware Pruning and Importance-Driven Token Merging for Efficient ViT","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28051","citing_title":"Beyond Surrogate Gradients: Fully Differentiable Token Pruning for Vision-Language Models","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19322","citing_title":"DynaTok: Temporally Adaptive and Positional Bias-Aware Token Compression for Video-LLMs","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05899","citing_title":"VisMMOE: Exploiting Visual-Expert Affinity for Efficient Visual-Language MoE Offloading","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04075","citing_title":"RetentiveKV: State-Space Memory for Uncertainty-Aware Multimodal KV Cache Eviction","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07812","citing_title":"HAWK: Head Importance-Aware Visual Token Pruning in Multimodal Models","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SZO3DK57JMPFAFFQXYMNT3IPPT","json":"https://pith.science/pith/SZO3DK57JMPFAFFQXYMNT3IPPT.json","graph_json":"https://pith.science/api/pith-number/SZO3DK57JMPFAFFQXYMNT3IPPT/graph.json","events_json":"https://pith.science/api/pith-number/SZO3DK57JMPFAFFQXYMNT3IPPT/events.json","paper":"https://pith.science/paper/SZO3DK57"},"agent_actions":{"view_html":"https://pith.science/pith/SZO3DK57JMPFAFFQXYMNT3IPPT","download_json":"https://pith.science/pith/SZO3DK57JMPFAFFQXYMNT3IPPT.json","view_paper":"https://pith.science/paper/SZO3DK57","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.11549&json=true","fetch_graph":"https://pith.science/api/pith-number/SZO3DK57JMPFAFFQXYMNT3IPPT/graph.json","fetch_events":"https://pith.science/api/pith-number/SZO3DK57JMPFAFFQXYMNT3IPPT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SZO3DK57JMPFAFFQXYMNT3IPPT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SZO3DK57JMPFAFFQXYMNT3IPPT/action/storage_attestation","attest_author":"https://pith.science/pith/SZO3DK57JMPFAFFQXYMNT3IPPT/action/author_attestation","sign_citation":"https://pith.science/pith/SZO3DK57JMPFAFFQXYMNT3IPPT/action/citation_signature","submit_replication":"https://pith.science/pith/SZO3DK57JMPFAFFQXYMNT3IPPT/action/replication_record"}},"created_at":"2026-07-05T10:31:34.300630+00:00","updated_at":"2026-07-05T10:31:34.300630+00:00"}