{"id":"3b25b1ff85d4d43265b522da9996d05fa80bfb65d293537f71c2cbefcabc80ab","title":"How Layer Deadlines Shape LLM Lookup Table Placement","argument":"Wen Huang and Jack Liu argue that where LLM lookup tables reside should depend on when each layer needs its data, as well as session reuse and contention on a shared device. Their system, SlackTier, starts table lookups at the beginning of a step, reads the layer with the shorter deadline first, and splits a small GPU row cache across layers. The authors report that Layer 14’s table can reside on NVMe without a cache and with no measurable decode cost. In their agent and chat replays, placing both tables on four NVMe disks raises mean time per output token by at most 1.4% over unmodified SGLang, while reducing host memory use from 212 to 34 GiB. They also report that RDMA reads from another node’s memory allow both tables to run without a cache. For a KV-constrained agent workload, they use the freed memory as host KV cache and report 36% higher median throughput and a reduction in median p90 time to first token from 38 to 24 seconds. These results connect table placement to a practical tradeoff: memory freed from lookup tables can support KV cache, but the reported performance depends on the workloads and measured read times.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence gives the paper’s abstract but not its methods or detailed results; the reported performance should therefore be read in the scope of the workloads and measurements described there.","speaker":{"name":"Huang, Wen, Liu, Jack","roleAtSource":"Article author"},"source":{"title":"Placing LLM Lookup Tables by Per-Layer Prefetch Deadlines","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.13413","publishedAt":"2026-11-10","dateBasis":"Date basis not provided."},"evidence":[{"text":"Huang and Liu argue that table placement depends on per-layer deadlines, session reuse, and contention on a shared device.","excerpt":"We show that this deadline, together with session reuse and contention on a shared device, decides which tier each table can use and which layers need cache","url":"https://arxiv.org/abs/2611.13413","relationship":"Source-stated"},{"text":"The authors describe SlackTier as issuing lookups at step start, prioritizing the short-window layer, and splitting a small GPU row cache across layers.","excerpt":"It reads the short-window layer first and splits a small GPU row cache across layers.","url":"https://arxiv.org/abs/2611.13413","relationship":"Source-stated"},{"text":"The authors report that Layer 14’s table can be placed on NVMe without a cache and with no measurable decode cost.","excerpt":"Layer 14's table can go to NVMe uncached with no measurable decode cost.","url":"https://arxiv.org/abs/2611.13413","relationship":"Source-stated"},{"text":"In agent and chat replays, the authors report at most a 1.4% increase in mean time per output token with both tables on four NVMe disks, while host memory falls from 212 to 34 GiB.","excerpt":"With both tables on four NVMe disks, mean time per output token on agent and chat replays rises at most 1.4% over unmodified SGLang, and host memory falls from 212 to 34 GiB.","url":"https://arxiv.org/abs/2611.13413","relationship":"Source-stated"},{"text":"For a KV-constrained agent workload, the authors report improved median throughput and lower median p90 time to first token when freed memory is used as host KV cache.","excerpt":"using the freed memory as host KV cache at the same memory budget raises median throughput by 36% and cuts median p90 time to first token from 38 to 24 s.","url":"https://arxiv.org/abs/2611.13413","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.13413","checkedAt":"2026-11-12T03:12:11.159Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How do the measured read times and workload conditions affect whether SlackTier’s reported results hold beyond the agent and chat replays described?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"How Layer Deadlines Shape LLM Lookup Table Placement","summary":"Wen Huang and Jack Liu argue that where LLM lookup tables reside should depend on when each layer needs its data, as well as session reuse and contention on a shared device. Their system, SlackTier, starts table lookups at the beginning of a step, reads the layer with the shorter deadline first, and splits a small GPU row cache across layers. The authors report that Layer 14’s table can reside on NVMe without a cache and with no measurable decode cost. In their agent and chat replays, placing both tables on four NVMe disks raises mean time per output token by at most 1.4% over unmodified SGLang, while reducing host memory use from 212 to 34 GiB. They also report that RDMA reads from another node’s memory allow both tables to run without a cache. For a KV-constrained agent workload, they use the freed memory as host KV cache and report 36% higher median throughput and a reduction in median p90 time to first token from 38 to 24 seconds. These results connect table placement to a practical tradeoff: memory freed from lookup tables can support KV cache, but the reported performance depends on the workloads and measured read times.","question":"How do the measured read times and workload conditions affect whether SlackTier’s reported results hold beyond the agent and chat replays described?","claims":[{"text":"Huang and Liu argue that table placement depends on per-layer deadlines, session reuse, and contention on a shared device.","excerpt":"We show that this deadline, together with session reuse and contention on a shared device, decides which tier each table can use and which layers need cache","support":"source-stated"},{"text":"The authors describe SlackTier as issuing lookups at step start, prioritizing the short-window layer, and splitting a small GPU row cache across layers.","excerpt":"It reads the short-window layer first and splits a small GPU row cache across layers.","support":"source-stated"},{"text":"The authors report that Layer 14’s table can be placed on NVMe without a cache and with no measurable decode cost.","excerpt":"Layer 14's table can go to NVMe uncached with no measurable decode cost.","support":"source-stated"},{"text":"In agent and chat replays, the authors report at most a 1.4% increase in mean time per output token with both tables on four NVMe disks, while host memory falls from 212 to 34 GiB.","excerpt":"With both tables on four NVMe disks, mean time per output token on agent and chat replays rises at most 1.4% over unmodified SGLang, and host memory falls from 212 to 34 GiB.","support":"source-stated"},{"text":"For a KV-constrained agent workload, the authors report improved median throughput and lower median p90 time to first token when freed memory is used as host KV cache.","excerpt":"using the freed memory as host KV cache at the same memory budget raises median throughput by 36% and cuts median p90 time to first token from 38 to 24 s.","support":"source-stated"}],"limitations":["The supplied evidence gives the paper’s abstract but not its methods or detailed results; the reported performance should therefore be read in the scope of the workloads and measurements described there."],"mediaId":"567bb049e254241ec587703ea98315e4412fdc9b07163ab223d99f1a054bc8b5","correctionReason":null,"id":"3b25b1ff85d4d43265b522da9996d05fa80bfb65d293537f71c2cbefcabc80ab","publicationId":"e5ae0632-7c46-4ce6-8a6f-97d72250d2a6","version":1,"sourceUrl":"https://arxiv.org/abs/2611.13413","sourceTitle":"Placing LLM Lookup Tables by Per-Layer Prefetch Deadlines","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-10","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-11T05:09:22.170Z","createdAt":"2026-11-12T01:05:06.982Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"567bb049e254241ec587703ea98315e4412fdc9b07163ab223d99f1a054bc8b5","kind":"article","url":"https://arxiv.org/abs/2611.13413","sourceUrl":"https://arxiv.org/abs/2611.13413","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T01:03:49.045Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.13413","checkedAt":"2026-11-12T03:12:11.159Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"567bb049e254241ec587703ea98315e4412fdc9b07163ab223d99f1a054bc8b5","confidence":0.71,"provider":"typesafe","model":"jev-1.13.0","inputHash":"e610702702c500502b65c4e77e6303720740821c686d7cb811de7837d36cdf5c","decidedAt":"2026-11-12T01:03:51.399Z"},"provenance":{"textHash":"754d68d83ab12368ecf8aed1f10e3ad484bea422733eb8c746fe3c6d49df4116","observationId":"b835d3bd2c111d3deb7ef3dea0bbef3c32c62a3c4e88127e7be7924c5f93e9c2","retrievedAt":"2026-11-12T01:03:48.741Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"e838ce0e-e9e2-4b4b-8dd0-05f20bbc5b20"},"reviewer":{"model":"gpt-6-luna","runId":"98d8d770-5aaf-4a8e-85a8-041a7a67817e"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-12T01:05:06.982Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Huang, Wen","url":null,"kind":"unknown"},{"name":"Liu, Jack","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Huang, Wen","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Wen","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13413","locator":"head/meta[name=author|citation_author]"},"assertionId":"889f29676bc8811a534155da4077e67816daf22f72118b5d5f247ae3ace215a1:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:12:11.159Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"889f29676bc8811a534155da4077e67816daf22f72118b5d5f247ae3ace215a1","sourceUrl":"https://arxiv.org/abs/2611.13413","contentSha256":"db3eace0ca8bbb9952e8a09114e5de7e62eb7b17548a9566ad9eee8f6327ddd0","textHash":"754d68d83ab12368ecf8aed1f10e3ad484bea422733eb8c746fe3c6d49df4116","observedAt":"2026-11-12T03:12:11.159Z"}},{"actor":{"name":"Liu, Jack","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Jack","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13413","locator":"head/meta[name=author|citation_author]"},"assertionId":"889f29676bc8811a534155da4077e67816daf22f72118b5d5f247ae3ace215a1:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:12:11.159Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"889f29676bc8811a534155da4077e67816daf22f72118b5d5f247ae3ace215a1","sourceUrl":"https://arxiv.org/abs/2611.13413","contentSha256":"db3eace0ca8bbb9952e8a09114e5de7e62eb7b17548a9566ad9eee8f6327ddd0","textHash":"754d68d83ab12368ecf8aed1f10e3ad484bea422733eb8c746fe3c6d49df4116","observedAt":"2026-11-12T03:12:11.159Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.13413","snapshotId":"889f29676bc8811a534155da4077e67816daf22f72118b5d5f247ae3ace215a1","contentSha256":"db3eace0ca8bbb9952e8a09114e5de7e62eb7b17548a9566ad9eee8f6327ddd0","textHash":"754d68d83ab12368ecf8aed1f10e3ad484bea422733eb8c746fe3c6d49df4116","observedAt":"2026-11-12T03:12:11.159Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/3b25b1ff85d4d43265b522da9996d05fa80bfb65d293537f71c2cbefcabc80ab/source"}},"canonicalUrl":"https://escapemeridian.com/readings/how-layer-deadlines-shape-llm-lookup-table-placement-3b25b1ff","formats":{"html":"/readings/how-layer-deadlines-shape-llm-lookup-table-placement-3b25b1ff","markdown":"/readings/how-layer-deadlines-shape-llm-lookup-table-placement-3b25b1ff.md","json":"/readings/how-layer-deadlines-shape-llm-lookup-table-placement-3b25b1ff.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-llm","name":"Large language model","explanation":"A language model with many learned numerical settings, used to process and generate language.","referenceUrl":"https://developers.google.com/machine-learning/glossary#large-language-model"},{"id":"term-gpu","name":"Graphics processing unit","explanation":"A chip that performs many calculations at once; the main hardware used to train and run AI models.","referenceUrl":"https://en.wikipedia.org/wiki/Graphics_processing_unit"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"},{"id":"time","name":"TIME","kind":"organization","context":"TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.","referenceUrls":["https://time.com/"],"path":"/entities/time","json":"/entities/time.json","markdown":"/entities/time.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"3b25b1ff85d4d43265b522da9996d05fa80bfb65d293537f71c2cbefcabc80ab","revision":"e5ae0632-7c46-4ce6-8a6f-97d72250d2a6","revisionKind":"publication-id","title":"How Layer Deadlines Shape LLM Lookup Table Placement","url":"https://escapemeridian.com/readings/how-layer-deadlines-shape-llm-lookup-table-placement-3b25b1ff"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.13413","revision":"754d68d83ab12368ecf8aed1f10e3ad484bea422733eb8c746fe3c6d49df4116","revisionKind":"sha256-source-text","title":"Placing LLM Lookup Tables by Per-Layer Prefetch Deadlines","url":"https://arxiv.org/abs/2611.13413"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"3b25b1ff85d4d43265b522da9996d05fa80bfb65d293537f71c2cbefcabc80ab","title":"How Layer Deadlines Shape LLM Lookup Table Placement","argument":"Wen Huang and Jack Liu argue that where LLM lookup tables reside should depend on when each layer needs its data, as well as session reuse and contention on a shared device. Their system, SlackTier, starts table lookups at the beginning of a step, reads the layer with the shorter deadline first, and splits a small GPU row cache across layers. The authors report that Layer 14’s table can reside on NVMe without a cache and with no measurable decode cost. In their agent and chat replays, placing both tables on four NVMe disks raises mean time per output token by at most 1.4% over unmodified SGLang, while reducing host memory use from 212 to 34 GiB. They also report that RDMA reads from another node’s memory allow both tables to run without a cache. For a KV-constrained agent workload, they use the freed memory as host KV cache and report 36% higher median throughput and a reduction in median p90 time to first token from 38 to 24 seconds. These results connect table placement to a practical tradeoff: memory freed from lookup tables can support KV cache, but the reported performance depends on the workloads and measured read times.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The supplied evidence gives the paper’s abstract but not its methods or detailed results; the reported performance should therefore be read in the scope of the workloads and measurements described there.","speaker":{"name":"Huang, Wen, Liu, Jack","roleAtSource":"Article author"},"source":{"title":"Placing LLM Lookup Tables by Per-Layer Prefetch Deadlines","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.13413","publishedAt":"2026-11-10","dateBasis":"Date basis not provided."},"evidence":[{"text":"Huang and Liu argue that table placement depends on per-layer deadlines, session reuse, and contention on a shared device.","excerpt":"We show that this deadline, together with session reuse and contention on a shared device, decides which tier each table can use and which layers need cache","url":"https://arxiv.org/abs/2611.13413","relationship":"Source-stated"},{"text":"The authors describe SlackTier as issuing lookups at step start, prioritizing the short-window layer, and splitting a small GPU row cache across layers.","excerpt":"It reads the short-window layer first and splits a small GPU row cache across layers.","url":"https://arxiv.org/abs/2611.13413","relationship":"Source-stated"},{"text":"The authors report that Layer 14’s table can be placed on NVMe without a cache and with no measurable decode cost.","excerpt":"Layer 14's table can go to NVMe uncached with no measurable decode cost.","url":"https://arxiv.org/abs/2611.13413","relationship":"Source-stated"},{"text":"In agent and chat replays, the authors report at most a 1.4% increase in mean time per output token with both tables on four NVMe disks, while host memory falls from 212 to 34 GiB.","excerpt":"With both tables on four NVMe disks, mean time per output token on agent and chat replays rises at most 1.4% over unmodified SGLang, and host memory falls from 212 to 34 GiB.","url":"https://arxiv.org/abs/2611.13413","relationship":"Source-stated"},{"text":"For a KV-constrained agent workload, the authors report improved median throughput and lower median p90 time to first token when freed memory is used as host KV cache.","excerpt":"using the freed memory as host KV cache at the same memory budget raises median throughput by 36% and cuts median p90 time to first token from 38 to 24 s.","url":"https://arxiv.org/abs/2611.13413","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.13413","checkedAt":"2026-11-12T03:12:11.159Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How do the measured read times and workload conditions affect whether SlackTier’s reported results hold beyond the agent and chat replays described?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"How Layer Deadlines Shape LLM Lookup Table Placement","summary":"Wen Huang and Jack Liu argue that where LLM lookup tables reside should depend on when each layer needs its data, as well as session reuse and contention on a shared device. Their system, SlackTier, starts table lookups at the beginning of a step, reads the layer with the shorter deadline first, and splits a small GPU row cache across layers. The authors report that Layer 14’s table can reside on NVMe without a cache and with no measurable decode cost. In their agent and chat replays, placing both tables on four NVMe disks raises mean time per output token by at most 1.4% over unmodified SGLang, while reducing host memory use from 212 to 34 GiB. They also report that RDMA reads from another node’s memory allow both tables to run without a cache. For a KV-constrained agent workload, they use the freed memory as host KV cache and report 36% higher median throughput and a reduction in median p90 time to first token from 38 to 24 seconds. These results connect table placement to a practical tradeoff: memory freed from lookup tables can support KV cache, but the reported performance depends on the workloads and measured read times.","question":"How do the measured read times and workload conditions affect whether SlackTier’s reported results hold beyond the agent and chat replays described?","claims":[{"text":"Huang and Liu argue that table placement depends on per-layer deadlines, session reuse, and contention on a shared device.","excerpt":"We show that this deadline, together with session reuse and contention on a shared device, decides which tier each table can use and which layers need cache","support":"source-stated"},{"text":"The authors describe SlackTier as issuing lookups at step start, prioritizing the short-window layer, and splitting a small GPU row cache across layers.","excerpt":"It reads the short-window layer first and splits a small GPU row cache across layers.","support":"source-stated"},{"text":"The authors report that Layer 14’s table can be placed on NVMe without a cache and with no measurable decode cost.","excerpt":"Layer 14's table can go to NVMe uncached with no measurable decode cost.","support":"source-stated"},{"text":"In agent and chat replays, the authors report at most a 1.4% increase in mean time per output token with both tables on four NVMe disks, while host memory falls from 212 to 34 GiB.","excerpt":"With both tables on four NVMe disks, mean time per output token on agent and chat replays rises at most 1.4% over unmodified SGLang, and host memory falls from 212 to 34 GiB.","support":"source-stated"},{"text":"For a KV-constrained agent workload, the authors report improved median throughput and lower median p90 time to first token when freed memory is used as host KV cache.","excerpt":"using the freed memory as host KV cache at the same memory budget raises median throughput by 36% and cuts median p90 time to first token from 38 to 24 s.","support":"source-stated"}],"limitations":["The supplied evidence gives the paper’s abstract but not its methods or detailed results; the reported performance should therefore be read in the scope of the workloads and measurements described there."],"mediaId":"567bb049e254241ec587703ea98315e4412fdc9b07163ab223d99f1a054bc8b5","correctionReason":null,"id":"3b25b1ff85d4d43265b522da9996d05fa80bfb65d293537f71c2cbefcabc80ab","publicationId":"e5ae0632-7c46-4ce6-8a6f-97d72250d2a6","version":1,"sourceUrl":"https://arxiv.org/abs/2611.13413","sourceTitle":"Placing LLM Lookup Tables by Per-Layer Prefetch Deadlines","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-10","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-11T05:09:22.170Z","createdAt":"2026-11-12T01:05:06.982Z","historical":false,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"567bb049e254241ec587703ea98315e4412fdc9b07163ab223d99f1a054bc8b5","kind":"article","url":"https://arxiv.org/abs/2611.13413","sourceUrl":"https://arxiv.org/abs/2611.13413","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-12T01:03:49.045Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.13413","checkedAt":"2026-11-12T03:12:11.159Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v6","mediaId":"567bb049e254241ec587703ea98315e4412fdc9b07163ab223d99f1a054bc8b5","confidence":0.71,"provider":"typesafe","model":"jev-1.13.0","inputHash":"e610702702c500502b65c4e77e6303720740821c686d7cb811de7837d36cdf5c","decidedAt":"2026-11-12T01:03:51.399Z"},"provenance":{"textHash":"754d68d83ab12368ecf8aed1f10e3ad484bea422733eb8c746fe3c6d49df4116","observationId":"b835d3bd2c111d3deb7ef3dea0bbef3c32c62a3c4e88127e7be7924c5f93e9c2","retrievedAt":"2026-11-12T01:03:48.741Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"e838ce0e-e9e2-4b4b-8dd0-05f20bbc5b20"},"reviewer":{"model":"gpt-6-luna","runId":"98d8d770-5aaf-4a8e-85a8-041a7a67817e"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-12T01:05:06.982Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Huang, Wen","url":null,"kind":"unknown"},{"name":"Liu, Jack","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Huang, Wen","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Huang, Wen","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13413","locator":"head/meta[name=author|citation_author]"},"assertionId":"889f29676bc8811a534155da4077e67816daf22f72118b5d5f247ae3ace215a1:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:12:11.159Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"889f29676bc8811a534155da4077e67816daf22f72118b5d5f247ae3ace215a1","sourceUrl":"https://arxiv.org/abs/2611.13413","contentSha256":"db3eace0ca8bbb9952e8a09114e5de7e62eb7b17548a9566ad9eee8f6327ddd0","textHash":"754d68d83ab12368ecf8aed1f10e3ad484bea422733eb8c746fe3c6d49df4116","observedAt":"2026-11-12T03:12:11.159Z"}},{"actor":{"name":"Liu, Jack","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Liu, Jack","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.13413","locator":"head/meta[name=author|citation_author]"},"assertionId":"889f29676bc8811a534155da4077e67816daf22f72118b5d5f247ae3ace215a1:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-12T03:12:11.159Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"889f29676bc8811a534155da4077e67816daf22f72118b5d5f247ae3ace215a1","sourceUrl":"https://arxiv.org/abs/2611.13413","contentSha256":"db3eace0ca8bbb9952e8a09114e5de7e62eb7b17548a9566ad9eee8f6327ddd0","textHash":"754d68d83ab12368ecf8aed1f10e3ad484bea422733eb8c746fe3c6d49df4116","observedAt":"2026-11-12T03:12:11.159Z"}}],"participationCoverage":{"total":2,"returned":2,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.13413","snapshotId":"889f29676bc8811a534155da4077e67816daf22f72118b5d5f247ae3ace215a1","contentSha256":"db3eace0ca8bbb9952e8a09114e5de7e62eb7b17548a9566ad9eee8f6327ddd0","textHash":"754d68d83ab12368ecf8aed1f10e3ad484bea422733eb8c746fe3c6d49df4116","observedAt":"2026-11-12T03:12:11.159Z","basis":"publisher-page-metadata","sourceLinks":42,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/3b25b1ff85d4d43265b522da9996d05fa80bfb65d293537f71c2cbefcabc80ab/source"}}}],"audits":[{"publicationId":"e5ae0632-7c46-4ce6-8a6f-97d72250d2a6","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes the argument and SlackTier results to authors Wen Huang and Jack Liu and keeps the results framed as their reported findings.","excerpt":"Authors:Wen Huang, Jack Liu"},{"dimension":"summary","status":"uncertain","explanation":"The figures match the abstract, but a general reader lacks essential context. The draft never says the tables are DeepSeek-V4.1's 203 GB Engram n-gram tables, which currently occupy most host DRAM. It also leaves central terms unexplained: lookup tables, NVMe, RDMA, KV cache, SGLang, decode, and p90 time to first token. RDMA is expanded in the source. The draft omits that one RDMA copy serves two instances. Smallest repair: add the DeepSeek-V4.1/Engram context and the 190-of-223 GiB memory problem, then use term tooltips for the jargon.","excerpt":"DeepSeek-V4.1's 203 GB of Engram n-gram tables sit in host DRAM today, taking 190 of the 223 GiB a 4-GPU serving process uses."},{"dimension":"date","status":"pass","explanation":"originalPublished 2026-11-10 matches the arXiv v1 submission date. It is not the later feed date.","excerpt":"[Submitted on 10 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim matches a verbatim abstract excerpt. The closing tradeoff stays within the source's scope: freed memory used as host KV cache, and a slowdown model based on measured read times. The question is open, not leading.","excerpt":"we give a model that predicts the resulting slowdown from measured read times"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, the most direct access to the source.","excerpt":null}],"createdAt":"2026-11-12T01:05:06.982Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}