{ "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": "2610.29302", "title": "Compressing Onboard Language Models for Space Robots under a Behavioral-Fidelity Requirement", "summary": "It replaces choosing one post-training quantization family, or accepting implicit PTQ behavior, with per-linear-module selection among AWQ, GPTQ, and AutoRound checkpoints using final next-token KL against a BF16 reference. The cost is offline compression-time search, multiple compatible checkpoints, BF16 reference forward passes, optional labeled task data, and continued configuration control of a static artifact whose unquantized embedding/output layers are 43-46% at 3 bits. It can fail when general-domain KL/perplexity gains do not transfer to the production command distribution, when target-adapted selection overfits small labels, or when mixed-checkpoint compatibility and unquantized layers erode the expected memory and fidelity benefits.", "rationale": "For a small team quantizing 3-8B models on one GPU, this is a feasible offline way to buy fidelity at fixed bits, but the claim is benchmark quality rather than proven production or flight-domain behavior.", "technique": "Mixed-family PTQ composition by output-KL selection", "family": "quantizacao", "recommendation": "testar", "published": "2026-10-21", "as_of": "2026-10-22", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2610.29302", "independent_implementations": 1, "total_implementations": 2, "stars": 4, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "It replaces choosing one post-training quantization family, or accepting implicit PTQ behavior, with per-linear-module selection among AWQ, GPTQ, and AutoRound checkpoints using final next-token KL against a BF16 reference. The cost is offline compression-time search, multiple compatible checkpoints, BF16 reference forward passes, optional labeled task data, and continued configuration control of a static artifact whose unquantized embedding/output layers are 43-46% at 3 bits. It can fail when general-domain KL/perplexity gains do not transfer to the production command distribution, when target-adapted selection overfits small labels, or when mixed-checkpoint compatibility and unquantized layers erode the expected memory and fidelity benefits.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.29302", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Composed AWQ/GPTQ/AutoRound models beat every single-family model on held-out perplexity in all six 3/4-bit settings; KL-scored selection beats representation scoring in all 12 paired runs, and target-adapted mode raises 3-bit pooled accuracy by 4.3 and 4.5 points on two Llama models.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2610.29302", "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": "" }