{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7JBIPLEPZ5HQ6QIGMC4AOX63H6","short_pith_number":"pith:7JBIPLEP","schema_version":"1.0","canonical_sha256":"fa4287ac8fcf4f0f410660b8075fdb3fbf49ab8d71f7d0cfce9adde3e43cf590","source":{"kind":"arxiv","id":"2505.21889","version":2},"attestation_state":"computed","paper":{"title":"EFIM: Efficient Serving of LLMs for Infilling Tasks with Improved KV Cache Reuse","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Cheater Lin, Feng Liu, Hande Dong, Nong Xiao, Tianyu Guo, Xianwei Zhang, Yichong Leng","submitted_at":"2025-05-28T02:07:03Z","abstract_excerpt":"Large language models (LLMs) are often used for infilling tasks, which involve predicting or generating missing information in a given text. These tasks typically require multiple interactions with similar context. To reduce the computation of repeated historical tokens, cross-request key-value (KV) cache reuse, a technique that stores and reuses intermediate computations, has become a crucial method in multi-round interactive services. However, in infilling tasks, the KV cache reuse is often hindered by the structure of the prompt format, which typically consists of a prefix and suffix relati"},"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":"2505.21889","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-05-28T02:07:03Z","cross_cats_sorted":[],"title_canon_sha256":"819f418961d12dbdd56b1d25c620495a02c32aea84c8254f2e278477ec632007","abstract_canon_sha256":"535698775bf458e7919d3813683a1d95c840d91a5a0b0091f1a77454319cc003"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:11:47.585451Z","signature_b64":"wTytgPrstXtGPUNfBYrSHZsDZ1VjwTrrq9N5JuSDb/IHBBIsogkB7QXlhfMHtfzDvv9Y/pbWB+D6bJZ9o7KrDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fa4287ac8fcf4f0f410660b8075fdb3fbf49ab8d71f7d0cfce9adde3e43cf590","last_reissued_at":"2026-07-05T11:11:47.584932Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:11:47.584932Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"EFIM: Efficient Serving of LLMs for Infilling Tasks with Improved KV Cache Reuse","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Cheater Lin, Feng Liu, Hande Dong, Nong Xiao, Tianyu Guo, Xianwei Zhang, Yichong Leng","submitted_at":"2025-05-28T02:07:03Z","abstract_excerpt":"Large language models (LLMs) are often used for infilling tasks, which involve predicting or generating missing information in a given text. These tasks typically require multiple interactions with similar context. To reduce the computation of repeated historical tokens, cross-request key-value (KV) cache reuse, a technique that stores and reuses intermediate computations, has become a crucial method in multi-round interactive services. However, in infilling tasks, the KV cache reuse is often hindered by the structure of the prompt format, which typically consists of a prefix and suffix relati"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.21889","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/2505.21889/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":"2505.21889","created_at":"2026-07-05T11:11:47.584992+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.21889v2","created_at":"2026-07-05T11:11:47.584992+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.21889","created_at":"2026-07-05T11:11:47.584992+00:00"},{"alias_kind":"pith_short_12","alias_value":"7JBIPLEPZ5HQ","created_at":"2026-07-05T11:11:47.584992+00:00"},{"alias_kind":"pith_short_16","alias_value":"7JBIPLEPZ5HQ6QIG","created_at":"2026-07-05T11:11:47.584992+00:00"},{"alias_kind":"pith_short_8","alias_value":"7JBIPLEP","created_at":"2026-07-05T11:11:47.584992+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.13097","citing_title":"Functional Cache Grafting: Robust and Rapid Code-Policy Synthesis for Embodied Agents","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7JBIPLEPZ5HQ6QIGMC4AOX63H6","json":"https://pith.science/pith/7JBIPLEPZ5HQ6QIGMC4AOX63H6.json","graph_json":"https://pith.science/api/pith-number/7JBIPLEPZ5HQ6QIGMC4AOX63H6/graph.json","events_json":"https://pith.science/api/pith-number/7JBIPLEPZ5HQ6QIGMC4AOX63H6/events.json","paper":"https://pith.science/paper/7JBIPLEP"},"agent_actions":{"view_html":"https://pith.science/pith/7JBIPLEPZ5HQ6QIGMC4AOX63H6","download_json":"https://pith.science/pith/7JBIPLEPZ5HQ6QIGMC4AOX63H6.json","view_paper":"https://pith.science/paper/7JBIPLEP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.21889&json=true","fetch_graph":"https://pith.science/api/pith-number/7JBIPLEPZ5HQ6QIGMC4AOX63H6/graph.json","fetch_events":"https://pith.science/api/pith-number/7JBIPLEPZ5HQ6QIGMC4AOX63H6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7JBIPLEPZ5HQ6QIGMC4AOX63H6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7JBIPLEPZ5HQ6QIGMC4AOX63H6/action/storage_attestation","attest_author":"https://pith.science/pith/7JBIPLEPZ5HQ6QIGMC4AOX63H6/action/author_attestation","sign_citation":"https://pith.science/pith/7JBIPLEPZ5HQ6QIGMC4AOX63H6/action/citation_signature","submit_replication":"https://pith.science/pith/7JBIPLEPZ5HQ6QIGMC4AOX63H6/action/replication_record"}},"created_at":"2026-07-05T11:11:47.584992+00:00","updated_at":"2026-07-05T11:11:47.584992+00:00"}