{"id":"c6e65f40e6911dce4242f238ce18c4e8af7e74e3545401eb7243460e58b22193","title":"Authors present DCKV for compressing the prompt KV cache","argument":"The authors present DCKV, a codec for compressing the KV cache—the part of a prompt’s state that hybrid language models store in their full-attention layers so it can be reused. They train its transform, quantizer and entropy models together, optimizing for both code length and changes to the frozen language model’s output.\n\nOn Qwen3.5-2B, the authors report storing a compressed token in more than 55 times fewer bytes than bf16, with a loss of at most 1.8 points on three long-context benchmarks. At more than 100 times fewer bytes, they report a loss of at most 3.6 points. These results make the tradeoff concrete: the reported compression levels are accompanied by losses on the benchmarks, rather than compression being presented as cost-free. The abstract does not establish how well the approach transfers beyond this model and these benchmarks.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported results in the abstract are for Qwen3.5-2B and three long-context benchmarks; it does not establish performance on other models or workloads.","speaker":{"name":"Maximov, Egor, Merkelov, Aleksei, Alekseev, Vasiliy, Krivorotov, Ivan, Beliaev, Valentin, Arsenin, Nikita, Bocharov, Dmitry, Goncharov, Aleksei","roleAtSource":"Article author"},"source":{"title":"DCKV: Learned Transform Coding of the Prompt KV Cache in Hybrid Language Models","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.36950","publishedAt":"2026-10-27","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors present DCKV as a codec for the KV cache, trained end to end.","excerpt":"We present DCKV, a differentiable codec for the KV cache and the first KV-cache codec that is trained end to end.","url":"https://arxiv.org/abs/2610.36950","relationship":"Source-stated"},{"text":"The authors optimize the transform, quantizer and entropy models together for code length and changes to the frozen language model's output.","excerpt":"Its transform, its quantizer and its entropy models, which are conditioned on the prompt tokens and on already decoded neighbouring positions, are optimized together against the frozen language model, both for the code length that the entropy models assign and for the change of the model's output.","url":"https://arxiv.org/abs/2610.36950","relationship":"Source-stated"},{"text":"On Qwen3.5-2B, the authors report more than 55 times fewer bytes than bf16 at a loss of at most 1.8 points on three long-context benchmarks.","excerpt":"On Qwen3.5-2B, DCKV stores a compressed token in more than 55 times fewer bytes than bf16 at a loss of at most 1.8 points on three long-context benchmarks","url":"https://arxiv.org/abs/2610.36950","relationship":"Source-stated"},{"text":"The abstract reports a higher compression level with a larger stated loss bound.","excerpt":"in more than 100 times fewer bytes at a loss of at most 3.6 points.","url":"https://arxiv.org/abs/2610.36950","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.36950","checkedAt":"2026-11-14T04:36:08.658Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Would DCKV show a similar compression-and-benchmark-loss tradeoff on other models and workloads?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Authors present DCKV for compressing the prompt KV cache","summary":"The authors present DCKV, a codec for compressing the KV cache—the part of a prompt’s state that hybrid language models store in their full-attention layers so it can be reused. They train its transform, quantizer and entropy models together, optimizing for both code length and changes to the frozen language model’s output.\n\nOn Qwen3.5-2B, the authors report storing a compressed token in more than 55 times fewer bytes than bf16, with a loss of at most 1.8 points on three long-context benchmarks. At more than 100 times fewer bytes, they report a loss of at most 3.6 points. These results make the tradeoff concrete: the reported compression levels are accompanied by losses on the benchmarks, rather than compression being presented as cost-free. The abstract does not establish how well the approach transfers beyond this model and these benchmarks.","question":"Would DCKV show a similar compression-and-benchmark-loss tradeoff on other models and workloads?","claims":[{"text":"The authors present DCKV as a codec for the KV cache, trained end to end.","excerpt":"We present DCKV, a differentiable codec for the KV cache and the first KV-cache codec that is trained end to end.","support":"source-stated"},{"text":"The authors optimize the transform, quantizer and entropy models together for code length and changes to the frozen language model's output.","excerpt":"Its transform, its quantizer and its entropy models, which are conditioned on the prompt tokens and on already decoded neighbouring positions, are optimized together against the frozen language model, both for the code length that the entropy models assign and for the change of the model's output.","support":"source-stated"},{"text":"On Qwen3.5-2B, the authors report more than 55 times fewer bytes than bf16 at a loss of at most 1.8 points on three long-context benchmarks.","excerpt":"On Qwen3.5-2B, DCKV stores a compressed token in more than 55 times fewer bytes than bf16 at a loss of at most 1.8 points on three long-context benchmarks","support":"source-stated"},{"text":"The abstract reports a higher compression level with a larger stated loss bound.","excerpt":"in more than 100 times fewer bytes at a loss of at most 3.6 points.","support":"source-stated"}],"limitations":["The reported results in the abstract are for Qwen3.5-2B and three long-context benchmarks; it does not establish performance on other models or workloads."],"mediaId":"a240d5f5975a689a2986b8be5ea044268c1504505e586df7fb4c31b479a3886c","correctionReason":null,"id":"c6e65f40e6911dce4242f238ce18c4e8af7e74e3545401eb7243460e58b22193","publicationId":"077b3d1e-da08-476e-816c-d106016b3b50","version":1,"sourceUrl":"https://arxiv.org/abs/2610.36950","sourceTitle":"DCKV: Learned Transform Coding of the Prompt KV Cache in Hybrid Language Models","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-27","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-28T07:10:38.633Z","createdAt":"2026-11-14T04:04:32.361Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"a240d5f5975a689a2986b8be5ea044268c1504505e586df7fb4c31b479a3886c","kind":"article","url":"https://arxiv.org/abs/2610.36950","sourceUrl":"https://arxiv.org/abs/2610.36950","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-30T21:03:11.039Z","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/2610.36950","checkedAt":"2026-11-14T04:36:08.658Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"a240d5f5975a689a2986b8be5ea044268c1504505e586df7fb4c31b479a3886c","confidence":0.89,"provider":"typesafe","model":"jev-1.13.0","inputHash":"d93c67d71a2ca4ebaa2368895f1aad2bf6d9f195892abc8333a3403ba4c0a0fa","decidedAt":"2026-10-30T21:03:12.899Z"},"provenance":{"textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observationId":"96889c01f40c04ac715a4a91c42b21d27b842fc1843da64e4dc3dddb6a21bba5","retrievedAt":"2026-10-30T21:03:10.749Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"4df5705b-9f9a-4a32-8ed4-6a75c5696f82"},"reviewer":{"model":"gpt-6-luna","runId":"54afdef3-c4fe-40b2-872d-9ebbcd56d02e"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-14T04:04:32.361Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Maximov, Egor","url":null,"kind":"unknown"},{"name":"Merkelov, Aleksei","url":null,"kind":"unknown"},{"name":"Alekseev, Vasiliy","url":null,"kind":"unknown"},{"name":"Krivorotov, Ivan","url":null,"kind":"unknown"},{"name":"Beliaev, Valentin","url":null,"kind":"unknown"},{"name":"Arsenin, Nikita","url":null,"kind":"unknown"},{"name":"Bocharov, Dmitry","url":null,"kind":"unknown"},{"name":"Goncharov, Aleksei","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Maximov, Egor","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Maximov, Egor","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Merkelov, Aleksei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Merkelov, Aleksei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Alekseev, Vasiliy","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Alekseev, Vasiliy","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Krivorotov, Ivan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Krivorotov, Ivan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Beliaev, Valentin","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Beliaev, Valentin","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Arsenin, Nikita","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Arsenin, Nikita","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Bocharov, Dmitry","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Bocharov, Dmitry","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Goncharov, Aleksei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Goncharov, Aleksei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:7","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}}],"participationCoverage":{"total":8,"returned":8,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.36950","snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z","basis":"publisher-page-metadata","sourceLinks":48,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/c6e65f40e6911dce4242f238ce18c4e8af7e74e3545401eb7243460e58b22193/source"}},"canonicalUrl":"https://escapemeridian.com/readings/authors-present-dckv-for-compressing-the-prompt-kv-cache-c6e65f40","formats":{"html":"/readings/authors-present-dckv-for-compressing-the-prompt-kv-cache-c6e65f40","markdown":"/readings/authors-present-dckv-for-compressing-the-prompt-kv-cache-c6e65f40.md","json":"/readings/authors-present-dckv-for-compressing-the-prompt-kv-cache-c6e65f40.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":[],"conditions":[],"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-benchmark","name":"Benchmark","explanation":"A defined set of tests used to compare systems. Results depend on the tasks and measures selected.","referenceUrl":"https://arxiv.org/abs/2211.09110"}],"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"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"c6e65f40e6911dce4242f238ce18c4e8af7e74e3545401eb7243460e58b22193","revision":"077b3d1e-da08-476e-816c-d106016b3b50","revisionKind":"publication-id","title":"Authors present DCKV for compressing the prompt KV cache","url":"https://escapemeridian.com/readings/authors-present-dckv-for-compressing-the-prompt-kv-cache-c6e65f40"},"source":{"type":"source","id":"https://arxiv.org/abs/2610.36950","revision":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","revisionKind":"sha256-source-text","title":"DCKV: Learned Transform Coding of the Prompt KV Cache in Hybrid Language Models","url":"https://arxiv.org/abs/2610.36950"},"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":"c6e65f40e6911dce4242f238ce18c4e8af7e74e3545401eb7243460e58b22193","title":"Authors present DCKV for compressing the prompt KV cache","argument":"The authors present DCKV, a codec for compressing the KV cache—the part of a prompt’s state that hybrid language models store in their full-attention layers so it can be reused. They train its transform, quantizer and entropy models together, optimizing for both code length and changes to the frozen language model’s output.\n\nOn Qwen3.5-2B, the authors report storing a compressed token in more than 55 times fewer bytes than bf16, with a loss of at most 1.8 points on three long-context benchmarks. At more than 100 times fewer bytes, they report a loss of at most 3.6 points. These results make the tradeoff concrete: the reported compression levels are accompanied by losses on the benchmarks, rather than compression being presented as cost-free. The abstract does not establish how well the approach transfers beyond this model and these benchmarks.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The reported results in the abstract are for Qwen3.5-2B and three long-context benchmarks; it does not establish performance on other models or workloads.","speaker":{"name":"Maximov, Egor, Merkelov, Aleksei, Alekseev, Vasiliy, Krivorotov, Ivan, Beliaev, Valentin, Arsenin, Nikita, Bocharov, Dmitry, Goncharov, Aleksei","roleAtSource":"Article author"},"source":{"title":"DCKV: Learned Transform Coding of the Prompt KV Cache in Hybrid Language Models","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2610.36950","publishedAt":"2026-10-27","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors present DCKV as a codec for the KV cache, trained end to end.","excerpt":"We present DCKV, a differentiable codec for the KV cache and the first KV-cache codec that is trained end to end.","url":"https://arxiv.org/abs/2610.36950","relationship":"Source-stated"},{"text":"The authors optimize the transform, quantizer and entropy models together for code length and changes to the frozen language model's output.","excerpt":"Its transform, its quantizer and its entropy models, which are conditioned on the prompt tokens and on already decoded neighbouring positions, are optimized together against the frozen language model, both for the code length that the entropy models assign and for the change of the model's output.","url":"https://arxiv.org/abs/2610.36950","relationship":"Source-stated"},{"text":"On Qwen3.5-2B, the authors report more than 55 times fewer bytes than bf16 at a loss of at most 1.8 points on three long-context benchmarks.","excerpt":"On Qwen3.5-2B, DCKV stores a compressed token in more than 55 times fewer bytes than bf16 at a loss of at most 1.8 points on three long-context benchmarks","url":"https://arxiv.org/abs/2610.36950","relationship":"Source-stated"},{"text":"The abstract reports a higher compression level with a larger stated loss bound.","excerpt":"in more than 100 times fewer bytes at a loss of at most 3.6 points.","url":"https://arxiv.org/abs/2610.36950","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2610.36950","checkedAt":"2026-11-14T04:36:08.658Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Would DCKV show a similar compression-and-benchmark-loss tradeoff on other models and workloads?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Authors present DCKV for compressing the prompt KV cache","summary":"The authors present DCKV, a codec for compressing the KV cache—the part of a prompt’s state that hybrid language models store in their full-attention layers so it can be reused. They train its transform, quantizer and entropy models together, optimizing for both code length and changes to the frozen language model’s output.\n\nOn Qwen3.5-2B, the authors report storing a compressed token in more than 55 times fewer bytes than bf16, with a loss of at most 1.8 points on three long-context benchmarks. At more than 100 times fewer bytes, they report a loss of at most 3.6 points. These results make the tradeoff concrete: the reported compression levels are accompanied by losses on the benchmarks, rather than compression being presented as cost-free. The abstract does not establish how well the approach transfers beyond this model and these benchmarks.","question":"Would DCKV show a similar compression-and-benchmark-loss tradeoff on other models and workloads?","claims":[{"text":"The authors present DCKV as a codec for the KV cache, trained end to end.","excerpt":"We present DCKV, a differentiable codec for the KV cache and the first KV-cache codec that is trained end to end.","support":"source-stated"},{"text":"The authors optimize the transform, quantizer and entropy models together for code length and changes to the frozen language model's output.","excerpt":"Its transform, its quantizer and its entropy models, which are conditioned on the prompt tokens and on already decoded neighbouring positions, are optimized together against the frozen language model, both for the code length that the entropy models assign and for the change of the model's output.","support":"source-stated"},{"text":"On Qwen3.5-2B, the authors report more than 55 times fewer bytes than bf16 at a loss of at most 1.8 points on three long-context benchmarks.","excerpt":"On Qwen3.5-2B, DCKV stores a compressed token in more than 55 times fewer bytes than bf16 at a loss of at most 1.8 points on three long-context benchmarks","support":"source-stated"},{"text":"The abstract reports a higher compression level with a larger stated loss bound.","excerpt":"in more than 100 times fewer bytes at a loss of at most 3.6 points.","support":"source-stated"}],"limitations":["The reported results in the abstract are for Qwen3.5-2B and three long-context benchmarks; it does not establish performance on other models or workloads."],"mediaId":"a240d5f5975a689a2986b8be5ea044268c1504505e586df7fb4c31b479a3886c","correctionReason":null,"id":"c6e65f40e6911dce4242f238ce18c4e8af7e74e3545401eb7243460e58b22193","publicationId":"077b3d1e-da08-476e-816c-d106016b3b50","version":1,"sourceUrl":"https://arxiv.org/abs/2610.36950","sourceTitle":"DCKV: Learned Transform Coding of the Prompt KV Cache in Hybrid Language Models","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-10-27","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-10-28T07:10:38.633Z","createdAt":"2026-11-14T04:04:32.361Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"a240d5f5975a689a2986b8be5ea044268c1504505e586df7fb4c31b479a3886c","kind":"article","url":"https://arxiv.org/abs/2610.36950","sourceUrl":"https://arxiv.org/abs/2610.36950","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-10-30T21:03:11.039Z","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/2610.36950","checkedAt":"2026-11-14T04:36:08.658Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"a240d5f5975a689a2986b8be5ea044268c1504505e586df7fb4c31b479a3886c","confidence":0.89,"provider":"typesafe","model":"jev-1.13.0","inputHash":"d93c67d71a2ca4ebaa2368895f1aad2bf6d9f195892abc8333a3403ba4c0a0fa","decidedAt":"2026-10-30T21:03:12.899Z"},"provenance":{"textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observationId":"96889c01f40c04ac715a4a91c42b21d27b842fc1843da64e4dc3dddb6a21bba5","retrievedAt":"2026-10-30T21:03:10.749Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"4df5705b-9f9a-4a32-8ed4-6a75c5696f82"},"reviewer":{"model":"gpt-6-luna","runId":"54afdef3-c4fe-40b2-872d-9ebbcd56d02e"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-14T04:04:32.361Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Maximov, Egor","url":null,"kind":"unknown"},{"name":"Merkelov, Aleksei","url":null,"kind":"unknown"},{"name":"Alekseev, Vasiliy","url":null,"kind":"unknown"},{"name":"Krivorotov, Ivan","url":null,"kind":"unknown"},{"name":"Beliaev, Valentin","url":null,"kind":"unknown"},{"name":"Arsenin, Nikita","url":null,"kind":"unknown"},{"name":"Bocharov, Dmitry","url":null,"kind":"unknown"},{"name":"Goncharov, Aleksei","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Maximov, Egor","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Maximov, Egor","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Merkelov, Aleksei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Merkelov, Aleksei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Alekseev, Vasiliy","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Alekseev, Vasiliy","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Krivorotov, Ivan","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Krivorotov, Ivan","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Beliaev, Valentin","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Beliaev, Valentin","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Arsenin, Nikita","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Arsenin, Nikita","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:5","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Bocharov, Dmitry","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Bocharov, Dmitry","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:6","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}},{"actor":{"name":"Goncharov, Aleksei","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Goncharov, Aleksei","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2610.36950","locator":"head/meta[name=author|citation_author]"},"assertionId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4:participation:7","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-14T04:36:08.658Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","sourceUrl":"https://arxiv.org/abs/2610.36950","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z"}}],"participationCoverage":{"total":8,"returned":8,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2610.36950","snapshotId":"b6a52ed8e7d383393527a75975b69e38110dc4ce19cabf76a9f29f1b1d6b40d4","contentSha256":"b1358c00717dc242e5e2326a2717419980c1c5a113a9665799b68d58ee8e3f81","textHash":"06c798824dd416eb0a00e59b8a4fba2d8f6240d1bb50e7a5d9e7b857620665a0","observedAt":"2026-11-14T04:36:08.658Z","basis":"publisher-page-metadata","sourceLinks":48,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/c6e65f40e6911dce4242f238ce18c4e8af7e74e3545401eb7243460e58b22193/source"}}}],"audits":[{"publicationId":"077b3d1e-da08-476e-816c-d106016b3b50","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"DCKV and its results are consistently attributed to the paper's authors (Maximov et al., arXiv). Results are framed as what the authors report, not as established fact.","excerpt":"We present DCKV, a differentiable codec for the KV cache"},{"dimension":"summary","status":"uncertain","explanation":"The reading is faithful but drops the source's stated motivation: the cache grows with the prompt and must be stored or moved to reuse it. Without that, a general reader cannot see why a smaller cache matters. Smallest repair: add that clause. The summary also omits the authors' claim that DCKV is the first KV-cache codec trained end to end. Central terms (codec, bf16, quantizer, entropy models) need term-library tooltips, especially bf16, which is the baseline for the 55x/100x figures. The title and question are fair and non-leading.","excerpt":"this cache is still the part of a prompt's state that grows with the prompt and that must be stored or moved to reuse it"},{"dimension":"date","status":"pass","explanation":"originalPublished 2026-10-27 matches the v1 submission date on arXiv. It is not the later feed date.","excerpt":"[Submitted on 27 Oct 2026]"},{"dimension":"support","status":"pass","explanation":"All claims match the abstract verbatim in substance: joint optimization against the frozen model, 55x fewer bytes than bf16 at a loss of at most 1.8 points, and 100x at a loss of at most 3.6 points on three long-context benchmarks. The tradeoff framing and the limitation to a single model and these benchmarks are appropriately conditional. No benefits are invented.","excerpt":"in more than 100 times fewer bytes at a loss of at most 3.6 points"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper, which is the most direct access to the source.","excerpt":null}],"createdAt":"2026-11-14T04:04:32.361Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}