{ "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": "2610.19026", "title": "Joint Subspace Sparse Attention for Efficient Long-Context Decoding", "summary": "JSSA replaces full-KV attention during long-context decoding with training-free token-level sparse selection scored in a learned shared low-dimensional query-key subspace, executed via a gather-free indexed-attention kernel compatible with the standard GQA KV-cache layout. The cost is integrating a custom kernel and computing the subspace projection per model, plus residual selection overhead; the subspace is derived from a closed-form spectral solution, so there is no training pipeline to maintain. It can fail if the low-dimensional scoring loses ranking structure on out-of-distribution workloads, if the kernel is unavailable or unoptimized for the reader's hardware (gains were demonstrated on a B200, not a 24 GB consumer GPU), or if competing sparse methods' generation-length inflation reappears on atypical prompts.", "rationale": "The method is training-free and directly reduces decoding-time KV-cache traffic, which suits constrained infrastructure, but the reported gains depend on a custom kernel validated on B200-class hardware and must be confirmed on the reader's specific GPU and workload before production adoption.", "technique": "Joint Subspace Sparse Attention (JSSA)", "family": "cache_kv", "recommendation": "testar", "published": "2026-10-13", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.19026", "independent_implementations": 0, "total_implementations": 1, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "JSSA replaces full-KV attention during long-context decoding with training-free token-level sparse selection scored in a learned shared low-dimensional query-key subspace, executed via a gather-free indexed-attention kernel compatible with the standard GQA KV-cache layout. The cost is integrating a custom kernel and computing the subspace projection per model, plus residual selection overhead; the subspace is derived from a closed-form spectral solution, so there is no training pipeline to maintain. It can fail if the low-dimensional scoring loses ranking structure on out-of-distribution workloads, if the kernel is unavailable or unoptimized for the reader's hardware (gains were demonstrated on a B200, not a 24 GB consumer GPU), or if competing sparse methods' generation-length inflation reappears on atypical prompts.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.19026", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Up to 6.43x decoding throughput versus full-KV attention at 128K context on an NVIDIA B200, while remaining close to full-KV accuracy and reaching 96-99% of model-native DSA throughput without training.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.19026", "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": "" }