{ "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": "2610.02953", "title": "SlimKV: Joint Token-Feature KV Cache Compression with Reconstruction-Free Beacon Attention", "summary": "SlimKV replaces the standard full-dimensional KV cache with low-rank latent beacon states that are attended to directly, avoiding the full-dimensional reconstruction that feature-wise compression methods normally require before applying RoPE. The cost is a low-rank-aware training phase for the beacon KV projections plus layer-adaptive rank allocation, which modifies the model weights rather than being a drop-in inference-time change, and some quality degradation at high compression ratios. It can fail on workloads where the training distribution differs from deployment, on very long contexts beyond the validated 128K range, or when the team cannot reproduce the required training procedure.", "rationale": "The method requires low-rank-aware training of the base model's KV projections, which exceeds the reader's no-foundation-model-training constraint, so it is worth tracking but not adoptable until pretrained SlimKV checkpoints or a training-free variant become available.", "technique": "Joint token-feature KV compression with beacon states", "family": "cache_kv", "recommendation": "observar", "published": "2026-10-02", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.02953", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "SlimKV replaces the standard full-dimensional KV cache with low-rank latent beacon states that are attended to directly, avoiding the full-dimensional reconstruction that feature-wise compression methods normally require before applying RoPE. The cost is a low-rank-aware training phase for the beacon KV projections plus layer-adaptive rank allocation, which modifies the model weights rather than being a drop-in inference-time change, and some quality degradation at high compression ratios. It can fail on workloads where the training distribution differs from deployment, on very long contexts beyond the validated 128K range, or when the team cannot reproduce the required training procedure.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.02953", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "SlimKV outperforms baselines at 16x/32x KV-cache compression, retains over 96% of the uncompressed model's LongBench score at 4x/8x, and achieves up to 7.34x attention speedup and 3.38x end-to-end decoding speedup at 128K context.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.02953", "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": "" }