{ "schema_version": 1, "builder_plan": { "scenario": "Long customer conversations exhaust memory or limit concurrent requests.", "change": "Change only the attention cache policy, after checking model and runtime support.", "baseline": "The same model with its ordinary attention cache.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Peak cache memory, long-context answer quality, first-token latency, and requests completed.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2610.28192", "title": "Offline Makes Online Better: Calibrated Cache Compression for Video Generation", "summary": "CALM replaces uniform-ratio token merging in KV-cache compression for autoregressive video generation with a fixed layer-head merging allocation, selected offline via an attention-output error proxy and applied at inference through dynamic programming under a given cache budget. The cost is a one-time offline calibration per model, plus dependence on the assumption that attention-output error predicts end-to-end video quality; inference overhead itself is minimal once the risk map is built. It can fail if the calibrated sensitivity map does not transfer to new prompts, model versions, or cache budgets far from those used in calibration, and the 21% LPIPS gain is reported against the method's own uniform-merging baseline, not against uncompressed generation.", "rationale": "It is directly deployable on a single-GPU inference setup and reduces KV-cache memory with a measurable quality gain, but the benefit must be validated on the reader's specific video model, prompts, and memory budget before production adoption.", "technique": "Calibrated layer-head KV-cache token merging", "family": "cache_kv", "recommendation": "testar", "published": "2026-10-20", "as_of": "2026-10-21", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.28192", "independent_implementations": 0, "total_implementations": 1, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "CALM replaces uniform-ratio token merging in KV-cache compression for autoregressive video generation with a fixed layer-head merging allocation, selected offline via an attention-output error proxy and applied at inference through dynamic programming under a given cache budget. The cost is a one-time offline calibration per model, plus dependence on the assumption that attention-output error predicts end-to-end video quality; inference overhead itself is minimal once the risk map is built. It can fail if the calibrated sensitivity map does not transfer to new prompts, model versions, or cache budgets far from those used in calibration, and the 21% LPIPS gain is reported against the method's own uniform-merging baseline, not against uncompressed generation.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.28192", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Reduces LPIPS to original videos by around 21% relative to uniform token merging at the same memory budget, with higher video-quality scores, across three autoregressive video models.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.28192", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [], "equations_status": "not_fetched", "core_kind": "", "equations_fetched_at": "", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }