{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:BHCDPFVP7NNE44TZWEYYDEAWPO","short_pith_number":"pith:BHCDPFVP","schema_version":"1.0","canonical_sha256":"09c43796affb5a4e7279b1318190167b82c5d0bdd9387e0f79d974525a5e4c37","source":{"kind":"arxiv","id":"2503.16428","version":1},"attestation_state":"computed","paper":{"title":"XAttention: Block Sparse Attention with Antidiagonal Scoring","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Guangxuan Xiao, Haofeng Huang, Junxian Guo, Ruyi Xu, Song Han","submitted_at":"2025-03-20T17:59:58Z","abstract_excerpt":"Long-Context Transformer Models (LCTMs) are vital for real-world applications but suffer high computational costs due to attention's quadratic complexity. Block-sparse attention mitigates this by focusing computation on critical regions, yet existing methods struggle with balancing accuracy and efficiency due to costly block importance measurements. In this paper, we introduce XAttention, a plug-and-play framework that dramatically accelerates long-context inference in Transformers models using sparse attention. XAttention's key innovation is the insight that the sum of antidiagonal values (i."},"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":"2503.16428","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-03-20T17:59:58Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"af3b23f0e68461d1c27ad18b47965d97857342e6ce4248e8df22538ebee2fc40","abstract_canon_sha256":"b98a0323e6b21397424df3137138a45a6115c2563e28fc75c816ccef44d992d4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:36:19.199289Z","signature_b64":"Obj99e5FQ3b4zCoqWZVGHkX19pKBojvVI1bfOE0Ug0XFNDfIZLTRYTr6UWBQLRm3m1KkidU/JCqlwBvQrPf+Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"09c43796affb5a4e7279b1318190167b82c5d0bdd9387e0f79d974525a5e4c37","last_reissued_at":"2026-07-05T10:36:19.198806Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:36:19.198806Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"XAttention: Block Sparse Attention with Antidiagonal Scoring","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Guangxuan Xiao, Haofeng Huang, Junxian Guo, Ruyi Xu, Song Han","submitted_at":"2025-03-20T17:59:58Z","abstract_excerpt":"Long-Context Transformer Models (LCTMs) are vital for real-world applications but suffer high computational costs due to attention's quadratic complexity. Block-sparse attention mitigates this by focusing computation on critical regions, yet existing methods struggle with balancing accuracy and efficiency due to costly block importance measurements. In this paper, we introduce XAttention, a plug-and-play framework that dramatically accelerates long-context inference in Transformers models using sparse attention. XAttention's key innovation is the insight that the sum of antidiagonal values (i."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.16428","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/2503.16428/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":"2503.16428","created_at":"2026-07-05T10:36:19.198865+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.16428v1","created_at":"2026-07-05T10:36:19.198865+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.16428","created_at":"2026-07-05T10:36:19.198865+00:00"},{"alias_kind":"pith_short_12","alias_value":"BHCDPFVP7NNE","created_at":"2026-07-05T10:36:19.198865+00:00"},{"alias_kind":"pith_short_16","alias_value":"BHCDPFVP7NNE44TZ","created_at":"2026-07-05T10:36:19.198865+00:00"},{"alias_kind":"pith_short_8","alias_value":"BHCDPFVP","created_at":"2026-07-05T10:36:19.198865+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":16,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.28283","citing_title":"PrunePath: Towards Highly Structured Sparse Language Models","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23445","citing_title":"DFSAttn: Dynamic Fine-grained Sparse Attention for Efficient Video Generation","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2510.18830","citing_title":"MTraining: Distributed Dynamic Sparse Attention for Efficient Ultra-Long Context Training","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2507.21526","citing_title":"Accelerating Prefilling via Decoding-time Contribution Sparsity","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2508.16703","citing_title":"ShadowNPU: System and Algorithm Co-design for NPU-Centric On-Device LLM Inference","ref_index":70,"is_internal_anchor":false},{"citing_arxiv_id":"2512.12602","citing_title":"Exact Flow Linear Attention: Exact Solution from Continuous-Time Dynamics","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2603.18636","citing_title":"Attention Sparsity is Input-Stable: Training-Free Sparse Attention for Video Generation via Offline Sparsity Profiling and Online QK Co-Clustering","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14513","citing_title":"HASTE: Training-Free Video Diffusion Acceleration via Head-Wise Adaptive Sparse Attention","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2510.26692","citing_title":"Kimi Linear: An Expressive, Efficient Attention Architecture","ref_index":112,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05888","citing_title":"MoE-Hub: Taming Software Complexity for Seamless MoE Overlap with Hardware-Accelerated Communication on Multi-GPU Systems","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21231","citing_title":"SparKV: Overhead-Aware KV Cache Loading for Efficient On-Device LLM Inference","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12219","citing_title":"Ride the Wave: Precision-Allocated Sparse Attention for Smooth Video Generation","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07719","citing_title":"An Efficient Hybrid Sparse Attention with CPU-GPU Parallelism for Long-Context Inference","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06763","citing_title":"Sparse Attention as a Range Searching Problem: Towards an Inference-Efficient Index for KV Cache","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17834","citing_title":"AsyncSparse: Accelerating Sparse Matrix-Matrix Multiplication on Asynchronous GPU Architectures","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15911","citing_title":"Efficient Video Diffusion Models: Advancements and Challenges","ref_index":156,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BHCDPFVP7NNE44TZWEYYDEAWPO","json":"https://pith.science/pith/BHCDPFVP7NNE44TZWEYYDEAWPO.json","graph_json":"https://pith.science/api/pith-number/BHCDPFVP7NNE44TZWEYYDEAWPO/graph.json","events_json":"https://pith.science/api/pith-number/BHCDPFVP7NNE44TZWEYYDEAWPO/events.json","paper":"https://pith.science/paper/BHCDPFVP"},"agent_actions":{"view_html":"https://pith.science/pith/BHCDPFVP7NNE44TZWEYYDEAWPO","download_json":"https://pith.science/pith/BHCDPFVP7NNE44TZWEYYDEAWPO.json","view_paper":"https://pith.science/paper/BHCDPFVP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.16428&json=true","fetch_graph":"https://pith.science/api/pith-number/BHCDPFVP7NNE44TZWEYYDEAWPO/graph.json","fetch_events":"https://pith.science/api/pith-number/BHCDPFVP7NNE44TZWEYYDEAWPO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BHCDPFVP7NNE44TZWEYYDEAWPO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BHCDPFVP7NNE44TZWEYYDEAWPO/action/storage_attestation","attest_author":"https://pith.science/pith/BHCDPFVP7NNE44TZWEYYDEAWPO/action/author_attestation","sign_citation":"https://pith.science/pith/BHCDPFVP7NNE44TZWEYYDEAWPO/action/citation_signature","submit_replication":"https://pith.science/pith/BHCDPFVP7NNE44TZWEYYDEAWPO/action/replication_record"}},"created_at":"2026-07-05T10:36:19.198865+00:00","updated_at":"2026-07-05T10:36:19.198865+00:00"}