{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SOI4C3TZ4M6UUHD5FX4U3MEYOD","short_pith_number":"pith:SOI4C3TZ","schema_version":"1.0","canonical_sha256":"9391c16e79e33d4a1c7d2df94db09870f2c129ec3bdefa523d0d0a1722cd5338","source":{"kind":"arxiv","id":"2503.16163","version":1},"attestation_state":"computed","paper":{"title":"SpeCache: Speculative Key-Value Caching for Efficient Generation of LLMs","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jing Han, Kai Han, Shibo Jie, Yehui Tang, Zhi-Hong Deng","submitted_at":"2025-03-20T14:01:56Z","abstract_excerpt":"Transformer-based large language models (LLMs) have already achieved remarkable results on long-text tasks, but the limited GPU memory (VRAM) resources struggle to accommodate the linearly growing demand for key-value (KV) cache as the sequence length increases, which has become a bottleneck for the application of LLMs on long sequences. Existing KV cache compression methods include eviction, merging, or quantization of the KV cache to reduce its size. However, compression results in irreversible information forgetting, potentially affecting the accuracy of subsequent decoding. In this paper, "},"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":"2503.16163","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/publicdomain/zero/1.0/","primary_cat":"cs.CL","submitted_at":"2025-03-20T14:01:56Z","cross_cats_sorted":[],"title_canon_sha256":"8809e5c6f194204063694e8bfa9633fdfdb1216dec2665990941ea45ec6bcc7c","abstract_canon_sha256":"b2aa46538cb02f127e3aa6e57b31e1798b24b4a460e2dd6b14340da029ffb589"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:36:06.945421Z","signature_b64":"3XuduwdPVfO7aWZAuRKEu/f/SiISRTyStm4pzpAvBtm97Iq2Dqj9rNTlGrqgR595YRodFWEY+NyQgeveczMmBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9391c16e79e33d4a1c7d2df94db09870f2c129ec3bdefa523d0d0a1722cd5338","last_reissued_at":"2026-07-05T10:36:06.944917Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:36:06.944917Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SpeCache: Speculative Key-Value Caching for Efficient Generation of LLMs","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Jing Han, Kai Han, Shibo Jie, Yehui Tang, Zhi-Hong Deng","submitted_at":"2025-03-20T14:01:56Z","abstract_excerpt":"Transformer-based large language models (LLMs) have already achieved remarkable results on long-text tasks, but the limited GPU memory (VRAM) resources struggle to accommodate the linearly growing demand for key-value (KV) cache as the sequence length increases, which has become a bottleneck for the application of LLMs on long sequences. Existing KV cache compression methods include eviction, merging, or quantization of the KV cache to reduce its size. However, compression results in irreversible information forgetting, potentially affecting the accuracy of subsequent decoding. In this paper, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.16163","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/2503.16163/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":"2503.16163","created_at":"2026-07-05T10:36:06.944982+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.16163v1","created_at":"2026-07-05T10:36:06.944982+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.16163","created_at":"2026-07-05T10:36:06.944982+00:00"},{"alias_kind":"pith_short_12","alias_value":"SOI4C3TZ4M6U","created_at":"2026-07-05T10:36:06.944982+00:00"},{"alias_kind":"pith_short_16","alias_value":"SOI4C3TZ4M6UUHD5","created_at":"2026-07-05T10:36:06.944982+00:00"},{"alias_kind":"pith_short_8","alias_value":"SOI4C3TZ","created_at":"2026-07-05T10:36:06.944982+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SOI4C3TZ4M6UUHD5FX4U3MEYOD","json":"https://pith.science/pith/SOI4C3TZ4M6UUHD5FX4U3MEYOD.json","graph_json":"https://pith.science/api/pith-number/SOI4C3TZ4M6UUHD5FX4U3MEYOD/graph.json","events_json":"https://pith.science/api/pith-number/SOI4C3TZ4M6UUHD5FX4U3MEYOD/events.json","paper":"https://pith.science/paper/SOI4C3TZ"},"agent_actions":{"view_html":"https://pith.science/pith/SOI4C3TZ4M6UUHD5FX4U3MEYOD","download_json":"https://pith.science/pith/SOI4C3TZ4M6UUHD5FX4U3MEYOD.json","view_paper":"https://pith.science/paper/SOI4C3TZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.16163&json=true","fetch_graph":"https://pith.science/api/pith-number/SOI4C3TZ4M6UUHD5FX4U3MEYOD/graph.json","fetch_events":"https://pith.science/api/pith-number/SOI4C3TZ4M6UUHD5FX4U3MEYOD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SOI4C3TZ4M6UUHD5FX4U3MEYOD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SOI4C3TZ4M6UUHD5FX4U3MEYOD/action/storage_attestation","attest_author":"https://pith.science/pith/SOI4C3TZ4M6UUHD5FX4U3MEYOD/action/author_attestation","sign_citation":"https://pith.science/pith/SOI4C3TZ4M6UUHD5FX4U3MEYOD/action/citation_signature","submit_replication":"https://pith.science/pith/SOI4C3TZ4M6UUHD5FX4U3MEYOD/action/replication_record"}},"created_at":"2026-07-05T10:36:06.944982+00:00","updated_at":"2026-07-05T10:36:06.944982+00:00"}