{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:L3PFNSLVI3POBIKBF2IV5TQKPI","short_pith_number":"pith:L3PFNSLV","schema_version":"1.0","canonical_sha256":"5ede56c97546dee0a1412e915ece0a7a021f26004481fc94523f608559534c22","source":{"kind":"arxiv","id":"2405.12532","version":2},"attestation_state":"computed","paper":{"title":"PyramidInfer: Pyramid KV Cache Compression for High-throughput LLM Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dongjie Yang, Hai Zhao, Shilin Zhang, Xiaodong Han, Yan Gao, Yao Hu","submitted_at":"2024-05-21T06:46:37Z","abstract_excerpt":"Large Language Models (LLMs) have shown remarkable comprehension abilities but face challenges in GPU memory usage during inference, hindering their scalability for real-time applications like chatbots. To accelerate inference, we store computed keys and values (KV cache) in the GPU memory. Existing methods study the KV cache compression to reduce memory by pruning the pre-computed KV cache. However, they neglect the inter-layer dependency between layers and huge memory consumption in pre-computation. To explore these deficiencies, we find that the number of crucial keys and values that influe"},"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":"2405.12532","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-05-21T06:46:37Z","cross_cats_sorted":[],"title_canon_sha256":"45c241aad9978b63a70fb3aef896480e9503ebe14ead062ec4cfa055a8b3c27d","abstract_canon_sha256":"2071930b197b676787b6fe6f1f90960b98069d596e3bd49eb5703f9cdf10b2dd"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:27:48.904963Z","signature_b64":"XOPzMnVeFBtaJszu5ZQpTOIlJ8q9EDJJsS0scG6+qxqnSzKli66DPCQXvKCFJOdE1rAXK/J6v8/QeiP1pHZmDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5ede56c97546dee0a1412e915ece0a7a021f26004481fc94523f608559534c22","last_reissued_at":"2026-07-05T08:27:48.904428Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:27:48.904428Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"PyramidInfer: Pyramid KV Cache Compression for High-throughput LLM Inference","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dongjie Yang, Hai Zhao, Shilin Zhang, Xiaodong Han, Yan Gao, Yao Hu","submitted_at":"2024-05-21T06:46:37Z","abstract_excerpt":"Large Language Models (LLMs) have shown remarkable comprehension abilities but face challenges in GPU memory usage during inference, hindering their scalability for real-time applications like chatbots. To accelerate inference, we store computed keys and values (KV cache) in the GPU memory. Existing methods study the KV cache compression to reduce memory by pruning the pre-computed KV cache. However, they neglect the inter-layer dependency between layers and huge memory consumption in pre-computation. To explore these deficiencies, we find that the number of crucial keys and values that influe"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.12532","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/2405.12532/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":"2405.12532","created_at":"2026-07-05T08:27:48.904479+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.12532v2","created_at":"2026-07-05T08:27:48.904479+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.12532","created_at":"2026-07-05T08:27:48.904479+00:00"},{"alias_kind":"pith_short_12","alias_value":"L3PFNSLVI3PO","created_at":"2026-07-05T08:27:48.904479+00:00"},{"alias_kind":"pith_short_16","alias_value":"L3PFNSLVI3POBIKB","created_at":"2026-07-05T08:27:48.904479+00:00"},{"alias_kind":"pith_short_8","alias_value":"L3PFNSLV","created_at":"2026-07-05T08:27:48.904479+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08032","citing_title":"What to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and Agents","ref_index":135,"is_internal_anchor":true},{"citing_arxiv_id":"2606.09508","citing_title":"From Rigid to Dynamic: Entropy-Guided Adaptive Inference for Long-Context LLMs","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00523","citing_title":"ProactiveLLM: Learning Active Interaction for Streaming Large Language Models","ref_index":112,"is_internal_anchor":false},{"citing_arxiv_id":"2410.13846","citing_title":"LightTransfer: Your Long-Context LLM is Secretly a Hybrid Model with Effortless Adaptation","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2502.01941","citing_title":"Semantic Integrity Matters: Benchmarking and Preserving High-Density Reasoning in KV Cache Compression","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18053","citing_title":"Protection Is (Nearly) All You Need: Structural Protection Dominates Scoring in Globally Capped KV Eviction","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2406.02069","citing_title":"PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08840","citing_title":"ReST-KV: Robust KV Cache Eviction with Layer-wise Output Reconstruction and Spatial-Temporal Smoothing","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23553","citing_title":"ClusterFusion++: Expanding Cluster-Level Fusion to Full Transformer-Block Decoding","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07234","citing_title":"Reformulating KV Cache Eviction Problem for Long-Context LLM Inference","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05219","citing_title":"Sparse Prefix Caching for Hybrid and Recurrent LLM Serving","ref_index":31,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/L3PFNSLVI3POBIKBF2IV5TQKPI","json":"https://pith.science/pith/L3PFNSLVI3POBIKBF2IV5TQKPI.json","graph_json":"https://pith.science/api/pith-number/L3PFNSLVI3POBIKBF2IV5TQKPI/graph.json","events_json":"https://pith.science/api/pith-number/L3PFNSLVI3POBIKBF2IV5TQKPI/events.json","paper":"https://pith.science/paper/L3PFNSLV"},"agent_actions":{"view_html":"https://pith.science/pith/L3PFNSLVI3POBIKBF2IV5TQKPI","download_json":"https://pith.science/pith/L3PFNSLVI3POBIKBF2IV5TQKPI.json","view_paper":"https://pith.science/paper/L3PFNSLV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.12532&json=true","fetch_graph":"https://pith.science/api/pith-number/L3PFNSLVI3POBIKBF2IV5TQKPI/graph.json","fetch_events":"https://pith.science/api/pith-number/L3PFNSLVI3POBIKBF2IV5TQKPI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/L3PFNSLVI3POBIKBF2IV5TQKPI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/L3PFNSLVI3POBIKBF2IV5TQKPI/action/storage_attestation","attest_author":"https://pith.science/pith/L3PFNSLVI3POBIKBF2IV5TQKPI/action/author_attestation","sign_citation":"https://pith.science/pith/L3PFNSLVI3POBIKBF2IV5TQKPI/action/citation_signature","submit_replication":"https://pith.science/pith/L3PFNSLVI3POBIKBF2IV5TQKPI/action/replication_record"}},"created_at":"2026-07-05T08:27:48.904479+00:00","updated_at":"2026-07-05T08:27:48.904479+00:00"}