{ "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": "2609.28974", "title": "Same Bit Width, Different Outcomes: Post-Training Quantization of Text-to-Speech Across Architectures", "summary": "The method replaces full-precision TTS weights with 4-bit per-channel quantization, using a staged ablation to locate the model-specific sensitive component and per-layer GPTQ to repair it. The cost is a quantization and validation pipeline per model and target runtime: outcomes at the same bit width vary by architecture (2.8 UTMOS loss on Supertonic versus 0.07 on Kokoro), so no single recipe transfers. What can fail is quality collapse from per-tensor scaling even at 8 bits, misidentified sensitive layers when skipping the ablation, and kernels that are slower than fp32 on a given runtime despite lower bit width.", "rationale": "PTQ plus GPTQ is feasible on a single 24 GB GPU and cuts serving cost, but the paper itself shows gains and failures are model- and hardware-specific, so adoption requires per-workload validation rather than blind rollout.", "technique": "Cross-architecture post-training quantization of TTS with staged ablation and per-layer GPTQ", "family": "quantizacao", "recommendation": "testar", "published": "2026-09-24", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.28974", "independent_implementations": 3, "total_implementations": 4, "stars": 674, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces full-precision TTS weights with 4-bit per-channel quantization, using a staged ablation to locate the model-specific sensitive component and per-layer GPTQ to repair it. The cost is a quantization and validation pipeline per model and target runtime: outcomes at the same bit width vary by architecture (2.8 UTMOS loss on Supertonic versus 0.07 on Kokoro), so no single recipe transfers. What can fail is quality collapse from per-tensor scaling even at 8 bits, misidentified sensitive layers when skipping the ablation, and kernels that are slower than fp32 on a given runtime despite lower bit width.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.28974", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "On a Mac mini, a 4-bit weight kernel runs Supertonic at 0.60x fp32 latency (roughly 1.67x faster), while int8 is slower; per-layer GPTQ restores quality to within 0.1 UTMOS of the unquantized model.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.28974", "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": "" }