{ "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": "2610.29229", "title": "Cache Locally, Recompute Selectively for Streaming Video Understanding", "summary": "FrameBlend replaces long-context idle-time prefill with short-window prefill plus query-time re-prefill of selected video tokens, whose recomputed KVs overwrite the stale cached entries. It costs an additional partial prefill pass at query time, adding latency and implementation complexity in token selection and cache replacement, though the paper reports this is outweighed by the prefill savings. It can fail when token selection misses query-relevant frames, when queries arrive too densely to amortize idle-time prefill, or with frameworks whose cache layouts do not support surgical KV replacement.", "rationale": "The method is training-free and plug-and-play on existing SVU frameworks, fitting a single-GPU budget, but its benefit is specific to streaming video workloads and requires validating token-selection quality and query-time latency on the reader's own traffic.", "technique": "Selective query-time KV recomputation for streaming video", "family": "cache_kv", "recommendation": "testar", "published": "2026-10-21", "as_of": "2026-10-22", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.29229", "independent_implementations": 1, "total_implementations": 2, "stars": 122, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "FrameBlend replaces long-context idle-time prefill with short-window prefill plus query-time re-prefill of selected video tokens, whose recomputed KVs overwrite the stale cached entries. It costs an additional partial prefill pass at query time, adding latency and implementation complexity in token selection and cache replacement, though the paper reports this is outweighed by the prefill savings. It can fail when token selection misses query-relevant frames, when queries arrive too densely to amortize idle-time prefill, or with frameworks whose cache layouts do not support surgical KV replacement.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.29229", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Over 20% reduction in per-frame prefill FLOPs on RVS-Ego and 14-59% lower total processing time for retrieval-based frameworks (including recomputation cost), plus accuracy gains up to 3.9 percentage points across three SVU frameworks and five benchmarks.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.29229", "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": "" }