{ "schema_version": 1, "builder_plan": { "scenario": "A self-hosted assistant is too expensive because its model uses most of the GPU memory.", "change": "Try the paper's supported lower-precision weights on a copy of your serving setup.", "baseline": "The same model at its original precision, on the same GPU.", "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 memory, task pass rate, latency, and total serving cost at the same context and load.", "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": "2608.18182", "title": "Efficient INT8 Inference of Small NLP Models on Server CPUs with PyTorch Native Stack", "summary": "Replaces FP32 CPU inference for BERT-family encoders with SmoothQuant INT8 through TorchAO, TorchInductor graph fusion, and oneDNN/AVX512_VNNI/AMX kernels. It requires calibration and recent PyTorch/TorchAO versions, with accuracy loss reported as negligible. Benefits depend on Xeon VNNI/AMX support and encoder workloads rather than generative LLMs, and calibration may not transfer to out-of-distribution data.", "rationale": "An upstream PyTorch/TorchAO path requiring no GPU or training, suitable for classification and retrieval on supported CPUs.", "technique": "SmoothQuant INT8 BERT encoders on CPU", "family": "quantizacao", "recommendation": "adotar", "published": "2026-08-18", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2608.18182", "independent_implementations": 0, "total_implementations": 1, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "Replaces FP32 CPU inference for BERT-family encoders with SmoothQuant INT8 through TorchAO, TorchInductor graph fusion, and oneDNN/AVX512_VNNI/AMX kernels. It requires calibration and recent PyTorch/TorchAO versions, with accuracy loss reported as negligible. Benefits depend on Xeon VNNI/AMX support and encoder workloads rather than generative LLMs, and calibration may not transfer to out-of-distribution data.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2608.18182", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Reports up to 5.8x end-to-end throughput speedup with negligible accuracy loss relative to the FP32 baseline.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2608.18182", "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": [ { "anchor": "S4.E3", "label": "(3)", "section": "4 Design and Implementation", "role": "proposed_method", "latex": "Y=X_{u8}\\cdot W^{\\top}-c.", "mathml": "Y=Xu​8⋅W⊤−c.", "context": "And the GEMM result is computed as", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" }, { "anchor": "S4.E1", "label": "(1)", "section": "4 Design and Implementation", "role": "proposed_method", "latex": "X_{u8}=\\operatorname{uint8}(\\operatorname{int32}(X_{s8})+128).", "mathml": "Xu​8=uint8⁡(int32⁡(Xs​8)+128).", "context": "", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" }, { "anchor": "S4.E2", "label": "(2)", "section": "4 Design and Implementation", "role": "proposed_method", "latex": "c=\\left(\\sum_{k=1}^{K}W_{:,k}\\right)\\cdot 128.", "mathml": "c=(∑k=1KW:,k)⋅128.", "context": "", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" } ], "equations_status": "selected", "core_kind": "formula", "equations_fetched_at": "2026-09-18", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }