{ "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": "2609.28029", "title": "Tensor Decomposition of Transformer Key-Value Caches: Spectral Structure and Format Comparison", "summary": "The method replaces a dense key-value cache with a four-way Tucker decomposition of the cache tensor, compressing only the low-rank token and feature modes while leaving heads and layers untouched. The cost is a reconstruction error that is higher for values than for keys at every ratio (values reach a higher error floor), plus decomposition and reconstruction overhead and the need to compress keys before RoPE, since post-RoPE keys lose 41-64% of their compressibility. It can fail because head and layer modes are nearly full-rank and resist compression, so effective ratios are capped, value reconstruction error may degrade generation quality, and 2D unfolding methods actually beat 4D Tucker for keys.", "rationale": "The spectral findings are directly actionable for a 24 GB GPU serving Mistral-7B or LLaMA-2-13B, but the paper reports only reconstruction error, so production adoption requires validating generation quality and latency on the actual workload.", "technique": "Tucker decomposition of the KV cache", "family": "cache_kv", "recommendation": "testar", "published": "2026-09-23", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.28029", "independent_implementations": 0, "total_implementations": 1, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces a dense key-value cache with a four-way Tucker decomposition of the cache tensor, compressing only the low-rank token and feature modes while leaving heads and layers untouched. The cost is a reconstruction error that is higher for values than for keys at every ratio (values reach a higher error floor), plus decomposition and reconstruction overhead and the need to compress keys before RoPE, since post-RoPE keys lose 41-64% of their compressibility. It can fail because head and layer modes are nearly full-rank and resist compression, so effective ratios are capped, value reconstruction error may degrade generation quality, and 2D unfolding methods actually beat 4D Tucker for keys.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.28029", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Tucker achieves the lowest reconstruction error at every compression ratio from 2x to 5x at matched storage, by leaving the full-rank head and layer modes uncompressed; no end-task quality or latency numbers are reported.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.28029", "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": "" }