Эксперимент: Certified dual-price MoE routing (#423)

{ "worked": true, "confidence": 8, "verdict": "Built a NumPy MVP for certified dual-price top-1 MoE routing, including projected expert-price updates, capacity enforcement, and the relaxed dual bound. Weak duality held in all 1,000 random trials, and the stability sweep showed bounded behavior for small rho and oscillatory collapse for large rho as predicted. In the matched toy comparison, rho=0.001 improved accepted utility, overflow, and load variance versus greedy routing, although the raw update can fail to rebalance discrete assignments and rho=0.1 was unstable.", "metrics": { "baseline": "Accepted utility 60.42; overflow tokens 63.70; load variance 1434.55", "idea": "rho=0.001: accepted utility 74.85; overflow tokens 10.40; load variance 56.47; mean dual gap 2.35. rho=0.01: utility 72.64; overflow 12.84; variance 66.31. rho=0.1: utility 21.85; overflow 95.75; variance 3058.69" }, "how_to_run": "/home/maxwelhelp/main/bin/python3 dual_moe_experiment.py", "files": [ "dual_moe_experiment.py", "results.json", "REPORT.md" ], "limitations": "No Transformer or language-model perplexity experiment was run; top-2 routing, EMA price smoothing, wall-clock latency, FLOPs, and GPU performance were not tested. The toy utilities and capacity process do not establish production MoE gains." }