{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:NVAFL2CYTUBAUAGXMEJWSSWGB6","short_pith_number":"pith:NVAFL2CY","schema_version":"1.0","canonical_sha256":"6d4055e8589d020a00d76113694ac60fa3a2eb5e7fd631754622ac45fd3f6da5","source":{"kind":"arxiv","id":"2501.01005","version":2},"attestation_state":"computed","paper":{"title":"FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"FlashInfer uses block-sparse KV-cache formats and JIT-compiled attention templates to cut inter-token latency by 29-69% in LLM serving.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.DC","authors_text":"Arvind Krishnamurthy, Baris Kasikci, Lequn Chen, Luis Ceze, Ruihang Lai, Stephanie Wang, Tianqi Chen, Vinod Grover, Wuwei Lin, Yineng Zhang, Zihao Ye","submitted_at":"2025-01-02T02:02:20Z","abstract_excerpt":"Transformers, driven by attention mechanisms, form the foundation of large language models (LLMs). As these models scale up, efficient GPU attention kernels become essential for high-throughput and low-latency inference. Diverse LLM applications demand flexible and high-performance attention solutions. We present FlashInfer: a customizable and efficient attention engine for LLM serving. FlashInfer tackles KV-cache storage heterogeneity using block-sparse format and composable formats to optimize memory access and reduce redundancy. It also offers a customizable attention template, enabling ada"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2501.01005","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.DC","submitted_at":"2025-01-02T02:02:20Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"fec089e8cc02be9f63d92004ce4011f481ba5effd987267046f165313ae1587b","abstract_canon_sha256":"ef8f3da93594914f409eba5625144ecd53a2459c3616175e48fa0c82a2e4b035"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:47.743715Z","signature_b64":"QoOhFvdckfGjafdkFh8Z9THlZcJhljkl1ltoAl1fR1BmUYV/jytGUrH7MDiOXxtr5DkZXQfjEQoN8lI/bQIJCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6d4055e8589d020a00d76113694ac60fa3a2eb5e7fd631754622ac45fd3f6da5","last_reissued_at":"2026-05-17T23:38:47.743252Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:47.743252Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FlashInfer: Efficient and Customizable Attention Engine for LLM Inference Serving","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"FlashInfer uses block-sparse KV-cache formats and JIT-compiled attention templates to cut inter-token latency by 29-69% in LLM serving.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.DC","authors_text":"Arvind Krishnamurthy, Baris Kasikci, Lequn Chen, Luis Ceze, Ruihang Lai, Stephanie Wang, Tianqi Chen, Vinod Grover, Wuwei Lin, Yineng Zhang, Zihao Ye","submitted_at":"2025-01-02T02:02:20Z","abstract_excerpt":"Transformers, driven by attention mechanisms, form the foundation of large language models (LLMs). As these models scale up, efficient GPU attention kernels become essential for high-throughput and low-latency inference. Diverse LLM applications demand flexible and high-performance attention solutions. We present FlashInfer: a customizable and efficient attention engine for LLM serving. FlashInfer tackles KV-cache storage heterogeneity using block-sparse format and composable formats to optimize memory access and reduce redundancy. It also offers a customizable attention template, enabling ada"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Compared to state-of-the-art LLM serving solutions, FlashInfer achieve 29-69% inter-token-latency reduction compared to compiler backends for LLM serving benchmark, 28-30% latency reduction for long-context inference, and 13-17% speedup for LLM serving with parallel generation.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The reported speedups assume that the block-sparse format and JIT templates integrate cleanly with existing serving frameworks without hidden overheads from compilation or scheduling that would appear under production load patterns not tested in the benchmarks.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"FlashInfer delivers a customizable attention kernel that reduces inter-token latency by 29-69% in LLM serving benchmarks via optimized KV-cache storage and load-balanced scheduling compatible with CUDA graphs.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"FlashInfer uses block-sparse KV-cache formats and JIT-compiled attention templates to cut inter-token latency by 29-69% in LLM serving.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b4b44bab60ff148d0db56ba6d96d2b1a261dac13863d61f6f6848b3b77d3d25b"},"source":{"id":"2501.01005","kind":"arxiv","version":2},"verdict":{"id":"d2941e40-5d8c-4a82-b667-2b1836ce3227","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T13:21:57.766174Z","strongest_claim":"Compared to state-of-the-art LLM serving solutions, FlashInfer achieve 29-69% inter-token-latency reduction compared to compiler backends for LLM serving benchmark, 28-30% latency reduction for long-context inference, and 13-17% speedup for LLM serving with parallel generation.","one_line_summary":"FlashInfer delivers a customizable attention kernel that reduces inter-token latency by 29-69% in LLM serving benchmarks via optimized KV-cache storage and load-balanced scheduling compatible with CUDA graphs.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The reported speedups assume that the block-sparse format and JIT templates integrate cleanly with existing serving frameworks without hidden overheads from compilation or scheduling that would appear under production load patterns not tested in the benchmarks.","pith_extraction_headline":"FlashInfer uses block-sparse KV-cache formats and JIT-compiled attention templates to cut inter-token latency by 29-69% in LLM serving."},"references":{"count":14,"sample":[{"doi":"10.1145/1583991.1584053","year":2004,"title":"URL https://arxiv.org/abs/2004. 05150. Buluç, A., Fineman, J. T., Frigo, M., Gilbert, J. R., and Leiserson, C. E. Parallel sparse matrix-vector and matrix-transpose-vector multiplication using com- pr","work_id":"4e1afef5-52c1-4ec8-8c63-7c4bf85ce428","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1145/3458817.3476182","year":2024,"title":"URL https: //doi.org/10.1145/3458817.3476182","work_id":"f104f07e-0962-463f-959e-ddecd7f1f7ba","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1177/1094342004041296","year":2024,"title":"org/paper_files/paper/2024/file/ 5321b1dabcd2be188d796c21b733e8c7- Paper-Conference.pdf","work_id":"db04265b-d798-47c9-b726-4ca1021628a1","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1109/sc41404.2022.00042","year":2022,"title":"El-rec: Efficient large-scale recommendation model training via tensor-train embedding table","work_id":"0625ddd0-436d-4c7a-b66a-97409cfc830c","ref_index":7,"cited_arxiv_id":"1404.2022","is_internal_anchor":true},{"doi":"10.1145/55364.55378","year":2024,"title":"URL https://doi","work_id":"a5dd3d03-6672-4aae-acf0-c19739d4a596","ref_index":9,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":14,"snapshot_sha256":"d5851aafbceae1d6888f616320a215806f2f3d523cfea2ed52d3d251580d9b49","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"1d4b34b6a74bdcbf05457dc3fcba8230a664874232639143fd07d4e35971cce2"},"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":"2501.01005","created_at":"2026-05-17T23:38:47.743330+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.01005v2","created_at":"2026-05-17T23:38:47.743330+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.01005","created_at":"2026-05-17T23:38:47.743330+00:00"},{"alias_kind":"pith_short_12","alias_value":"NVAFL2CYTUBA","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"NVAFL2CYTUBAUAGX","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"NVAFL2CY","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":32,"internal_anchor_count":32,"sample":[{"citing_arxiv_id":"2412.03594","citing_title":"BatchLLM: Optimizing Large Batched LLM Inference with Global Prefix Sharing and Throughput-oriented Token Batching","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2502.01068","citing_title":"FastKV: Decoupling of Context Reduction and KV Cache Compression for Prefill-Decoding Acceleration","ref_index":33,"is_internal_anchor":true},{"citing_arxiv_id":"2511.02043","citing_title":"Flashlight: PyTorch Compiler Extensions to Accelerate Attention Variants","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07985","citing_title":"Dooly: Configuration-Agnostic, Redundancy-Aware Profiling for LLM Inference Simulation","ref_index":40,"is_internal_anchor":true},{"citing_arxiv_id":"2605.22416","citing_title":"Asymmetric Virtual Memory Paging for Hybrid Mamba-Transformer Inference","ref_index":18,"is_internal_anchor":true},{"citing_arxiv_id":"2512.19179","citing_title":"CascadeInfer: Length-Aware Scheduling of LLM Serving with Low Latency and Load Balancing","ref_index":31,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18856","citing_title":"SPHERICAL KV: Angle-Domain Attention and Rate-Distortion Retention for Efficient Long-Context Inference","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2605.17633","citing_title":"SparseSAM: Structured Sparsification of Activations in Segment Anything Models","ref_index":32,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18071","citing_title":"KVDrive: A Holistic Multi-Tier KV Cache Management System for Long-Context LLM Inference","ref_index":38,"is_internal_anchor":true},{"citing_arxiv_id":"2510.08726","citing_title":"Neptune: Advanced ML Operator Fusion for Locality and Parallelism on GPUs","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2510.09883","citing_title":"DELTA: Dynamic Layer-Aware Token Attention for Efficient Long-Context Reasoning","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2510.18245","citing_title":"Scaling Laws Meet Model Architecture: Toward Inference-Efficient LLMs","ref_index":48,"is_internal_anchor":true},{"citing_arxiv_id":"2512.12087","citing_title":"BLASST: Dynamic BLocked Attention Sparsity via Softmax Thresholding","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2601.13684","citing_title":"HeteroCache: A Dynamic Retrieval Approach to Heterogeneous KV Cache Compression for Long-Context LLM Inference","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2601.14910","citing_title":"PipeWeave: Synergizing Analytical and Learning Models for Unified GPU Performance Prediction","ref_index":75,"is_internal_anchor":true},{"citing_arxiv_id":"2604.09558","citing_title":"VTC: DNN Compilation with Virtual Tensors for Data Movement Elimination","ref_index":36,"is_internal_anchor":true},{"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":14,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10195","citing_title":"Breaking the Reward Barrier: Accelerating Tree-of-Thought Reasoning via Speculative Exploration","ref_index":65,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00831","citing_title":"GhostServe: A Lightweight Checkpointing System in the Shadow for Fault-Tolerant LLM Serving","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2604.27476","citing_title":"EdgeFM: Efficient Edge Inference for Vision-Language Models","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10195","citing_title":"Breaking the Reward Barrier: Accelerating Tree-of-Thought Reasoning via Speculative Exploration","ref_index":65,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08467","citing_title":"CUDAHercules: Benchmarking Hardware-Aware Expert-level CUDA Optimization for LLMs","ref_index":26,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08524","citing_title":"Unleashing Scalable Context Parallelism for Foundation Models Pre-Training via FCP","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06068","citing_title":"VibeServe: Can AI Agents Build Bespoke LLM Serving Systems?","ref_index":80,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00616","citing_title":"LLM-Emu: Native Runtime Emulation of LLM Inference via Profile-Driven Sampling","ref_index":11,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NVAFL2CYTUBAUAGXMEJWSSWGB6","json":"https://pith.science/pith/NVAFL2CYTUBAUAGXMEJWSSWGB6.json","graph_json":"https://pith.science/api/pith-number/NVAFL2CYTUBAUAGXMEJWSSWGB6/graph.json","events_json":"https://pith.science/api/pith-number/NVAFL2CYTUBAUAGXMEJWSSWGB6/events.json","paper":"https://pith.science/paper/NVAFL2CY"},"agent_actions":{"view_html":"https://pith.science/pith/NVAFL2CYTUBAUAGXMEJWSSWGB6","download_json":"https://pith.science/pith/NVAFL2CYTUBAUAGXMEJWSSWGB6.json","view_paper":"https://pith.science/paper/NVAFL2CY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.01005&json=true","fetch_graph":"https://pith.science/api/pith-number/NVAFL2CYTUBAUAGXMEJWSSWGB6/graph.json","fetch_events":"https://pith.science/api/pith-number/NVAFL2CYTUBAUAGXMEJWSSWGB6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NVAFL2CYTUBAUAGXMEJWSSWGB6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NVAFL2CYTUBAUAGXMEJWSSWGB6/action/storage_attestation","attest_author":"https://pith.science/pith/NVAFL2CYTUBAUAGXMEJWSSWGB6/action/author_attestation","sign_citation":"https://pith.science/pith/NVAFL2CYTUBAUAGXMEJWSSWGB6/action/citation_signature","submit_replication":"https://pith.science/pith/NVAFL2CYTUBAUAGXMEJWSSWGB6/action/replication_record"}},"created_at":"2026-05-17T23:38:47.743330+00:00","updated_at":"2026-05-17T23:38:47.743330+00:00"}