{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:X2DYZKX6E46BCXB6FSKCL4KJ56","short_pith_number":"pith:X2DYZKX6","canonical_record":{"source":{"id":"2605.09735","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AR","submitted_at":"2026-05-10T20:10:26Z","cross_cats_sorted":["cs.AI","cs.DC","cs.OS"],"title_canon_sha256":"427ea78dbfd0d9626f6cb9a82146ca9b60d839a19e556005b4b1975e93ddb676","abstract_canon_sha256":"aa7f3762003a7b04d3c4d9a650fbfb482457e4056087999c301f99cb77d8b407"},"schema_version":"1.0"},"canonical_sha256":"be878caafe273c115c3e2c9425f149efb72a427012c874501668c7d439fbf225","source":{"kind":"arxiv","id":"2605.09735","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.09735","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"arxiv_version","alias_value":"2605.09735v2","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.09735","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_12","alias_value":"X2DYZKX6E46B","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_16","alias_value":"X2DYZKX6E46BCXB6","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_8","alias_value":"X2DYZKX6","created_at":"2026-07-01T01:17:15Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:X2DYZKX6E46BCXB6FSKCL4KJ56","target":"record","payload":{"canonical_record":{"source":{"id":"2605.09735","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AR","submitted_at":"2026-05-10T20:10:26Z","cross_cats_sorted":["cs.AI","cs.DC","cs.OS"],"title_canon_sha256":"427ea78dbfd0d9626f6cb9a82146ca9b60d839a19e556005b4b1975e93ddb676","abstract_canon_sha256":"aa7f3762003a7b04d3c4d9a650fbfb482457e4056087999c301f99cb77d8b407"},"schema_version":"1.0"},"canonical_sha256":"be878caafe273c115c3e2c9425f149efb72a427012c874501668c7d439fbf225","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-01T01:17:15.721659Z","signature_b64":"ksyvxnd+QI9xE48MInFX9bTe94VlRejnxHTSvFZ0Rdcvn1+hjRVT9DVVVUwR1YW5ZawN+xrmU86ztOuJ/KhKCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"be878caafe273c115c3e2c9425f149efb72a427012c874501668c7d439fbf225","last_reissued_at":"2026-07-01T01:17:15.721076Z","signature_status":"signed_v1","first_computed_at":"2026-07-01T01:17:15.721076Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.09735","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-01T01:17:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"0HE0l3yO/qXe8JGkUgVdlsVlW5mHfkqtPQ8oICvPByOSZiJKWIMfFmdyqZcLv3QqJgne/ovwtRqbNeqKL6/GAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-13T01:36:25.280684Z"},"content_sha256":"05bdbf95770eb767852ca66f750452e8a948861d2b4c2c9f9aa7a5705b9c9bf9","schema_version":"1.0","event_id":"sha256:05bdbf95770eb767852ca66f750452e8a948861d2b4c2c9f9aa7a5705b9c9bf9"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:X2DYZKX6E46BCXB6FSKCL4KJ56","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"KV-RM: Regularizing KV-Cache Movement for Static-Graph LLM Serving","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Regularizing KV-cache movement lets static-graph LLM decoders absorb variable request lengths without over-reserving memory.","cross_cats":["cs.AI","cs.DC","cs.OS"],"primary_cat":"cs.AR","authors_text":"Bolun Sun, Jian Zhang, Weijian Zheng, Xiaodong Yu, Zhijing Ye, Zhiqing Zhong","submitted_at":"2026-05-10T20:10:26Z","abstract_excerpt":"Static-graph LLM decoders provide predictable launches, fixed tensor shapes, and low submission overhead, but online decoding exposes highly irregular KV-cache behavior: request lengths differ, EOS events arrive asynchronously, and logical histories fragment over time. Dynamic runtimes recover flexibility through paged KV management and step-level scheduling, while static-graph executors often over-reserve memory and suffer burst-time latency outliers. This paper studies whether much of this variability can be absorbed below a fixed decode interface. We present KV-RM, a runtime design that reg"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On a 2-GPU NVIDIA A100 node, KV-RM improves mixed-length decoding throughput and tail latency relative to a static-graph baseline, reduces reserved KV memory across workload families, and removes severe burst-time latency spikes under production-trace replay.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the variability from differing request lengths, asynchronous EOS events, and fragmented histories can be absorbed below a fixed decode interface primarily through KV-cache movement regularization, without the design depending on the optional bounded far-history summaries.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"KV-RM regularizes KV-cache movement in static-graph LLM serving via block paging and merge-staged transport to improve throughput, tail latency, and memory use for variable-length decoding.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Regularizing KV-cache movement lets static-graph LLM decoders absorb variable request lengths without over-reserving memory.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"cea47200f676c6232eb11c49dc36c26c513a90249d3af444b1d4fe0681f3cc01"},"source":{"id":"2605.09735","kind":"arxiv","version":2},"verdict":{"id":"595f8a3e-255e-42ef-a2d2-2e1c8b1eca65","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T03:50:47.463862Z","strongest_claim":"On a 2-GPU NVIDIA A100 node, KV-RM improves mixed-length decoding throughput and tail latency relative to a static-graph baseline, reduces reserved KV memory across workload families, and removes severe burst-time latency spikes under production-trace replay.","one_line_summary":"KV-RM regularizes KV-cache movement in static-graph LLM serving via block paging and merge-staged transport to improve throughput, tail latency, and memory use for variable-length decoding.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the variability from differing request lengths, asynchronous EOS events, and fragmented histories can be absorbed below a fixed decode interface primarily through KV-cache movement regularization, without the design depending on the optional bounded far-history summaries.","pith_extraction_headline":"Regularizing KV-cache movement lets static-graph LLM decoders absorb variable request lengths without over-reserving memory."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.09735/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T07:22:01.301432Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T16:37:08.581355Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T12:31:18.094911Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T09:59:06.208418Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"0306149e2c18aadce5ccabe345138fa035dbedd8ec598ac568cefde2cfb43fed"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":3,"snapshot_sha256":"33e3a53ec94f39a15dfad6d00a518ceed674c1740e3557b23e8ca5fc528c0242"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"595f8a3e-255e-42ef-a2d2-2e1c8b1eca65"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-01T01:17:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"defuLnakLViopulQnMJ2vCuqyZBjsDCLxPkDPXdqsX5KFdV+7+draMqYFvXogvjuq+C7g8bSABkmamdiNLhPBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-13T01:36:25.281179Z"},"content_sha256":"4d0789e89b17071041c16092dbf19de20dafae449ba0d239a1d21664256c8432","schema_version":"1.0","event_id":"sha256:4d0789e89b17071041c16092dbf19de20dafae449ba0d239a1d21664256c8432"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/X2DYZKX6E46BCXB6FSKCL4KJ56/bundle.json","state_url":"https://pith.science/pith/X2DYZKX6E46BCXB6FSKCL4KJ56/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/X2DYZKX6E46BCXB6FSKCL4KJ56/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-13T01:36:25Z","links":{"resolver":"https://pith.science/pith/X2DYZKX6E46BCXB6FSKCL4KJ56","bundle":"https://pith.science/pith/X2DYZKX6E46BCXB6FSKCL4KJ56/bundle.json","state":"https://pith.science/pith/X2DYZKX6E46BCXB6FSKCL4KJ56/state.json","well_known_bundle":"https://pith.science/.well-known/pith/X2DYZKX6E46BCXB6FSKCL4KJ56/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:X2DYZKX6E46BCXB6FSKCL4KJ56","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"aa7f3762003a7b04d3c4d9a650fbfb482457e4056087999c301f99cb77d8b407","cross_cats_sorted":["cs.AI","cs.DC","cs.OS"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AR","submitted_at":"2026-05-10T20:10:26Z","title_canon_sha256":"427ea78dbfd0d9626f6cb9a82146ca9b60d839a19e556005b4b1975e93ddb676"},"schema_version":"1.0","source":{"id":"2605.09735","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.09735","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"arxiv_version","alias_value":"2605.09735v2","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.09735","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_12","alias_value":"X2DYZKX6E46B","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_16","alias_value":"X2DYZKX6E46BCXB6","created_at":"2026-07-01T01:17:15Z"},{"alias_kind":"pith_short_8","alias_value":"X2DYZKX6","created_at":"2026-07-01T01:17:15Z"}],"graph_snapshots":[{"event_id":"sha256:4d0789e89b17071041c16092dbf19de20dafae449ba0d239a1d21664256c8432","target":"graph","created_at":"2026-07-01T01:17:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"On a 2-GPU NVIDIA A100 node, KV-RM improves mixed-length decoding throughput and tail latency relative to a static-graph baseline, reduces reserved KV memory across workload families, and removes severe burst-time latency spikes under production-trace replay."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the variability from differing request lengths, asynchronous EOS events, and fragmented histories can be absorbed below a fixed decode interface primarily through KV-cache movement regularization, without the design depending on the optional bounded far-history summaries."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"KV-RM regularizes KV-cache movement in static-graph LLM serving via block paging and merge-staged transport to improve throughput, tail latency, and memory use for variable-length decoding."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Regularizing KV-cache movement lets static-graph LLM decoders absorb variable request lengths without over-reserving memory."}],"snapshot_sha256":"cea47200f676c6232eb11c49dc36c26c513a90249d3af444b1d4fe0681f3cc01"},"formal_canon":{"evidence_count":3,"snapshot_sha256":"33e3a53ec94f39a15dfad6d00a518ceed674c1740e3557b23e8ca5fc528c0242"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T07:22:01.301432Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T16:37:08.581355Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T12:31:18.094911Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T09:59:06.208418Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.09735/integrity.json","findings":[],"snapshot_sha256":"0306149e2c18aadce5ccabe345138fa035dbedd8ec598ac568cefde2cfb43fed","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Static-graph LLM decoders provide predictable launches, fixed tensor shapes, and low submission overhead, but online decoding exposes highly irregular KV-cache behavior: request lengths differ, EOS events arrive asynchronously, and logical histories fragment over time. Dynamic runtimes recover flexibility through paged KV management and step-level scheduling, while static-graph executors often over-reserve memory and suffer burst-time latency outliers. This paper studies whether much of this variability can be absorbed below a fixed decode interface. We present KV-RM, a runtime design that reg","authors_text":"Bolun Sun, Jian Zhang, Weijian Zheng, Xiaodong Yu, Zhijing Ye, Zhiqing Zhong","cross_cats":["cs.AI","cs.DC","cs.OS"],"headline":"Regularizing KV-cache movement lets static-graph LLM decoders absorb variable request lengths without over-reserving memory.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AR","submitted_at":"2026-05-10T20:10:26Z","title":"KV-RM: Regularizing KV-Cache Movement for Static-Graph LLM Serving"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.09735","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-12T03:50:47.463862Z","id":"595f8a3e-255e-42ef-a2d2-2e1c8b1eca65","model_set":{"reader":"grok-4.3"},"one_line_summary":"KV-RM regularizes KV-cache movement in static-graph LLM serving via block paging and merge-staged transport to improve throughput, tail latency, and memory use for variable-length decoding.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Regularizing KV-cache movement lets static-graph LLM decoders absorb variable request lengths without over-reserving memory.","strongest_claim":"On a 2-GPU NVIDIA A100 node, KV-RM improves mixed-length decoding throughput and tail latency relative to a static-graph baseline, reduces reserved KV memory across workload families, and removes severe burst-time latency spikes under production-trace replay.","weakest_assumption":"That the variability from differing request lengths, asynchronous EOS events, and fragmented histories can be absorbed below a fixed decode interface primarily through KV-cache movement regularization, without the design depending on the optional bounded far-history summaries."}},"verdict_id":"595f8a3e-255e-42ef-a2d2-2e1c8b1eca65"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:05bdbf95770eb767852ca66f750452e8a948861d2b4c2c9f9aa7a5705b9c9bf9","target":"record","created_at":"2026-07-01T01:17:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"aa7f3762003a7b04d3c4d9a650fbfb482457e4056087999c301f99cb77d8b407","cross_cats_sorted":["cs.AI","cs.DC","cs.OS"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AR","submitted_at":"2026-05-10T20:10:26Z","title_canon_sha256":"427ea78dbfd0d9626f6cb9a82146ca9b60d839a19e556005b4b1975e93ddb676"},"schema_version":"1.0","source":{"id":"2605.09735","kind":"arxiv","version":2}},"canonical_sha256":"be878caafe273c115c3e2c9425f149efb72a427012c874501668c7d439fbf225","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"be878caafe273c115c3e2c9425f149efb72a427012c874501668c7d439fbf225","first_computed_at":"2026-07-01T01:17:15.721076Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-01T01:17:15.721076Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"ksyvxnd+QI9xE48MInFX9bTe94VlRejnxHTSvFZ0Rdcvn1+hjRVT9DVVVUwR1YW5ZawN+xrmU86ztOuJ/KhKCg==","signature_status":"signed_v1","signed_at":"2026-07-01T01:17:15.721659Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.09735","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:05bdbf95770eb767852ca66f750452e8a948861d2b4c2c9f9aa7a5705b9c9bf9","sha256:4d0789e89b17071041c16092dbf19de20dafae449ba0d239a1d21664256c8432"],"state_sha256":"fd01ab8bccd34c05a3c571162e316860592f64efc139b7927c4fe6ebcd4dfec0"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"FzX0NAwZhis1wsQ7U745MaVLQoOhy5M05vZDta74gdy1j2vR3LmZxqYlNP+Idz8WaiOrl0l+6bbYh4+AxGcDDA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-13T01:36:25.284217Z","bundle_sha256":"383e58ecf599fdc83c66567af077eae64ddbdee9956bbbb6f9f08db427e17872"}}