{ "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": "2608.24945", "title": "FAMPWQ: Fisher Information-based Adaptive Mixed Precision Weight Quantization for Effective LLM Inference", "summary": "Replaces uniform bit-width or heuristic-sensitivity weight quantization with layer-wise sensitivity measured by Fisher information and a reinforcement learning allocator that assigns per-layer bit-widths. Costs include a Fisher sensitivity estimation pass, training an RL allocator, and added deployment complexity from mixed-precision kernels, which may offset some memory savings relative to uniform low-bit formats. Can fail if Fisher estimates misrank sensitivity on the reader's domain data, if commodity GPU kernels lack efficient mixed-precision support (turning memory savings into latency regressions), or if the allocator's strategy does not transfer across models.", "rationale": "Mixed-precision quantization fits a single 24 GB GPU and can preserve quality better than uniform low-bit schemes, but the RL allocator overhead and kernel support requirements require workload-specific validation before production adoption.", "technique": "Fisher-guided RL mixed-precision weight quantization", "family": "quantizacao", "recommendation": "testar", "published": "2026-08-24", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2608.24945", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "Replaces uniform bit-width or heuristic-sensitivity weight quantization with layer-wise sensitivity measured by Fisher information and a reinforcement learning allocator that assigns per-layer bit-widths. Costs include a Fisher sensitivity estimation pass, training an RL allocator, and added deployment complexity from mixed-precision kernels, which may offset some memory savings relative to uniform low-bit formats. Can fail if Fisher estimates misrank sensitivity on the reader's domain data, if commodity GPU kernels lack efficient mixed-precision support (turning memory savings into latency regressions), or if the allocator's strategy does not transfer across models.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2608.24945", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Outperforms 7 baselines with up to 3.39 lower perplexity, up to 6.87% higher accuracy, and up to 76% LLM-as-a-judge win rate across 7 models and 5 benchmarks; no multiplicative speed or memory factor is stated.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2608.24945", "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.E11", "label": "(11)", "section": "4 FAMPWQ Methodology", "role": "loss", "latex": "\\min\\mathcal{L}(\\mathcal{Q})=\\Delta Acc(\\mathcal{Q})+P(\\psi(\\mathcal{Q}))\\cdot||\\psi(\\mathcal{Q})||^{2},", "mathml": "min⁡ℒ⁡(𝒬)=Δ​A​c​c​(𝒬)+P⁡(ψ⁡(𝒬))⋅‖ψ⁡(𝒬)‖2,", "context": "In this section, we present an RL-based adaptive bit-width allocation method. Since the combinatorial problem defined in Formula 4 is intractable, we transform it into a single loss function minimization problem as defined in Formula 11.", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" }, { "anchor": "S4.E10", "label": "(10)", "section": "4 FAMPWQ Methodology", "role": "metric", "latex": "{s}_{l}=\\frac{\\left\\|\\mathcal{F}(\\theta_{l}^{\\text{pert}})-\\mathcal{F}(\\theta_{l})\\right\\|_{2}}{\\left\\|\\mathcal{F}(\\theta_{l})\\right\\|_{2}},", "mathml": "sl=‖ℱ⁡(θlpert)−ℱ⁡(θl)‖2‖ℱ⁡(θl)‖2,", "context": "Finally, we take the FIM variation to indicate the sensitivity of a layer, which is defined in Formula 10.", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" }, { "anchor": "S3.E4", "label": "(4)", "section": "3 System Model and Problem Formulation", "role": "proposed_method", "latex": "\\mathcal{Q}^{*}=\\operatorname*{argmin}_{\\mathcal{Q}}\\Delta Acc(\\mathcal{Q}), \\\\ \\text{s.t.} \\begin{cases}\\forall\\quad(l,q_{l})\\in\\mathcal{Q}^{*},q_{l}\\in\\mathcal{B},\\\\\n\\sum_{l=1}^{L}p_{l}q_{l}\\leq R\\cdot B\\cdot\\sum_{l=1}^{L}p_{l},\\end{cases}", "mathml": "𝒬∗=argmin𝒬Δ​A​c​c​(𝒬),s.t.{∀(l,ql)∈𝒬∗,ql∈ℬ,∑l=1Lpl​ql≤R⋅B⋅∑l=1Lpl,", "context": "The problem we address in this work is how to find a bit-width allocation strategy 𝒬∗ so as to minimize the accuracy degradation while achieving the compression rate target as formulated in Formula 4.", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" } ], "equations_status": "selected", "core_kind": "formula", "equations_fetched_at": "2026-09-21", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }