{ "schema_version": 1, "builder_plan": { "scenario": "Your inference service slows down when several customers send requests together.", "change": "Replay a recorded request trace with only the proposed serving policy changed.", "baseline": "Your current scheduler on the same model and hardware.", "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": "Requests meeting your latency target, throughput, failures, and cost per successful request.", "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": "2610.29873", "title": "HAWK: Heterogeneity-Aware KV Cache Transfer for Efficient GPU-NPU Phase-Disaggregated LLM Inference", "summary": "HAWK replaces single-device-type serving with phase-disaggregated inference in which prefill runs on GPUs and decode on NPUs, connected by an RDMA path that remaps KV cache tensor layouts into NPU-side layout-aware buffers during transfer. It costs dedicated heterogeneous hardware (commercial NPUs plus RDMA-capable interconnect), custom transfer orchestration, and up to 7.88% throughput loss relative to a GPU-only baseline. It can fail through cross-device layout mismatches, RDMA path or interconnect instability, and engineering fragility when models, KV formats, or hardware generations change.", "rationale": "The method presupposes co-located GPU and NPU hardware with RDMA interconnect that the reader's single 24 GB GPU or API-based environment cannot provide, so there is no production action to take.", "technique": "Heterogeneous GPU-NPU phase disaggregation with layout-aware KV transfer", "family": "serving_e_batching", "recommendation": "nao_aplica", "published": "2026-10-21", "as_of": "2026-10-22", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.29873", "independent_implementations": 0, "total_implementations": 1, "stars": 0, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "HAWK replaces single-device-type serving with phase-disaggregated inference in which prefill runs on GPUs and decode on NPUs, connected by an RDMA path that remaps KV cache tensor layouts into NPU-side layout-aware buffers during transfer. It costs dedicated heterogeneous hardware (commercial NPUs plus RDMA-capable interconnect), custom transfer orchestration, and up to 7.88% throughput loss relative to a GPU-only baseline. It can fail through cross-device layout mismatches, RDMA path or interconnect instability, and engineering fragility when models, KV formats, or hardware generations change.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.29873", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Achieves up to 1.72x higher cost efficiency and 1.26x higher power efficiency than a GPU-only system, with throughput within 7.88% of GPU-only under the same SLO constraints.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.29873", "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": "" }