{ "schema_version": 1, "builder_plan": { "scenario": "You already need task-specific training and want to reduce its memory or compute cost.", "change": "Change only the training technique on a small controlled run after checking prerequisites.", "baseline": "Your current training recipe with the same model, data split, and evaluation.", "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": "Held-out quality, peak memory, training time, and full compute cost to the target quality.", "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": "2609.24678", "title": "Muon Can Outperform Dedicated Continual Learning Methods", "summary": "Replaces loss-based orthogonality penalties in dedicated continual-learning LoRA methods (O-LoRA, ELLA) with plain incremental LoRA trained under the Muon optimizer, which orthogonalizes each update. Cost is limited to swapping the optimizer for LoRA parameters, with no extra memory, latency, or loss terms, but stacking both constraint mechanisms can cost 8.4 accuracy points and reduce plasticity. Results rest on two benchmarks with limited seeds and task orders, so the benefit may not transfer to different model scales, adapter configurations, or task sequences, and Muon's behavior on non-LoRA parameters still requires AdamW-style handling.", "rationale": "The optimizer swap is cheap and compatible with a single 24 GB GPU, but the evidence comes from specific CL benchmarks, so the reader should validate it on their own task sequence before production adoption.", "technique": "Muon-orthogonalized incremental LoRA for continual learning", "family": "treino_eficiente", "recommendation": "testar", "published": "2026-09-21", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.24678", "independent_implementations": 0, "total_implementations": 0, "stars": 0, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "Replaces loss-based orthogonality penalties in dedicated continual-learning LoRA methods (O-LoRA, ELLA) with plain incremental LoRA trained under the Muon optimizer, which orthogonalizes each update. Cost is limited to swapping the optimizer for LoRA parameters, with no extra memory, latency, or loss terms, but stacking both constraint mechanisms can cost 8.4 accuracy points and reduce plasticity. Results rest on two benchmarks with limited seeds and task orders, so the benefit may not transfer to different model scales, adapter configurations, or task sequences, and Muon's behavior on non-LoRA parameters still requires AdamW-style handling.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.24678", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "IncLoRA+Muon reaches the accuracy band of dedicated CL methods (O-LoRA, ELLA) on the Standard CL Benchmark, improves on every AdamW configuration on TRACE, and outperforms the most restrictive combined setup by 8.4 accuracy points; gains attributed to spreading updates over ~7.0 effective singular directions versus 1.4-1.8 under AdamW.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.24678", "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": "" }