{ "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": "2609.18110", "title": "SSD-LLaMA: SSD-Native Inference for Trillion-Parameter MoE at 1+ Token/s on a Consumer PC", "summary": "It replaces RAM/VRAM-resident expert storage (or expert pruning and substitution) with an SSD-native three-tier hierarchy that streams and retains experts dynamically across SSD, RAM, and VRAM with CPU-GPU hybrid execution. The cost is substantial systems complexity: an SSD I/O pipeline, expert caching and scheduling across three tiers, and dependence on fast NVMe storage, plus decode rates around 1 token/s that are unsuitable for interactive serving. It can fail when expert reuse is low and workloads become SSD-bandwidth-bound, when the available GPU is weaker than the evaluated RTX 5090, or when per-token expert selection defeats the caching policy, and unvalidated I/O stalls may dominate latency.", "rationale": "The system targets self-hosting trillion-parameter MoE models on a specific 32 GB GPU plus NVMe setup that the reader does not have, and third-party APIs already provide these models without the engineering burden, so it is worth tracking but not adopting.", "technique": "SSD-native tiered expert offloading for MoE inference", "family": "serving_e_batching", "recommendation": "observar", "published": "2026-09-16", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.18110", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "It replaces RAM/VRAM-resident expert storage (or expert pruning and substitution) with an SSD-native three-tier hierarchy that streams and retains experts dynamically across SSD, RAM, and VRAM with CPU-GPU hybrid execution. The cost is substantial systems complexity: an SSD I/O pipeline, expert caching and scheduling across three tiers, and dependence on fast NVMe storage, plus decode rates around 1 token/s that are unsuitable for interactive serving. It can fail when expert reuse is low and workloads become SSD-bandwidth-bound, when the available GPU is weaker than the evaluated RTX 5090, or when per-token expert selection defeats the caching policy, and unvalidated I/O stalls may dominate latency.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.18110", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Improves prefill token rate by 1.52x-4.19x and decode token rate by 2.10x-15.58x over baselines across three frontier MoE families; sustains above 1 token/s on a trillion-parameter model with a single RTX 5090 and at most 32 GB RAM.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.18110", "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": "" }