Стенд-проверка (stage-2) · промт оператора:

(универсальный)

Ответ агента:

{ "worked": false, "confidence": 10, "verdict": "The registered custom track correlated_token_moe_regression was benchmarked end-to-end with eight paired seeds. Shell-wise balancing exactly achieved zero measured load CV and zero per-shell spread, but the best test MSE improvement from 0.0566415 to 0.0542420 was not significant (paired permutation p=0.38035), so the idea did not produce the required significant win.", "metrics": { "baseline": "Registered-track baseline, best lr=0.006 and auxiliary loss=0.01; full 8-seed mean test MSE 0.056641495786607265; mean load CV 0.11686608614250746; mean shell spread 258.375.", "idea": "Registered-track shell-balanced MoE, best lr=0.006 and K=8; full 8-seed mean test MSE 0.05424201488494873; mean load CV 0.0; mean shell spread 0.0; mean reassigned fraction 0.503828125.", "paired_delta": "-0.002399480901658535, 5/8 idea wins, permutation p=0.38035" }, "bench_report": { "bench_version": 1, "track": "correlated_token_moe_regression", "model": "custom_moe", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "aux": 0.01, "balanced": false, "K": 4 }, "sweep": [ { "cfg": { "lr": 0.0015, "aux": 0.0, "balanced": false, "K": 4 }, "mean": 0.0866615679115057 }, { "cfg": { "lr": 0.0015, "aux": 0.01, "balanced": false, "K": 4 }, "mean": 0.08026662468910217 }, { "cfg": { "lr": 0.0015, "aux": 0.1, "balanced": false, "K": 4 }, "mean": 0.079 }, { "cfg": { "lr": 0.003, "aux": 0.0, "balanced": false, "K": 4 }, "mean": 0.071 }, { "cfg": { "lr": 0.003, "aux": 0.01, "balanced": false, "K": 4 }, "mean": 0.064 }, { "cfg": { "lr": 0.003, "aux": 0.1, "balanced": false, "K": 4 }, "mean": 0.064 }, { "cfg": { "lr": 0.006, "aux": 0.0, "balanced": false, "K": 4 }, "mean": 0.058 }, { "cfg": { "lr": 0.006, "aux": 0.01, "balanced": false, "K": 4 }, "mean": 0.056641495786607265 }, { "cfg": { "lr": 0.006, "aux": 0.1, "balanced": false, "K": 4 }, "mean": 0.057 } ], "full": { "per_seed": [ 0.04945246428251267, 0.058252088725566864, 0.05263408645987511, 0.06455036252737045, 0.053144052624702454, 0.07924238234758377, 0.04448141157627106, 0.05137510327959061 ], "mean": 0.056641495786607265, "cfg": { "lr": 0.006, "aux": 0.01, "balanced": false, "K": 4 } } }, "idea": { "per_seed": [ 0.05467335134744644, 0.05477738380432129, 0.056872498244047165, 0.051517780870199203, 0.05167350545525551, 0.06755945831537247, 0.04719924181699753, 0.04966289922595024 ], "mean": 0.05424201488494873, "cfg": { "lr": 0.006, "balanced": true, "K": 8 } }, "comparison": { "delta_mean": -0.002399480901658535, "idea_wins": 5, "n_pairs": 8, "per_seed_diffs": [ 0.005220886319875717, -0.0034751594066619873, 0.004238411784172058, -0.013032834976911545, -0.0014705471694469452, -0.0116824209690094, 0.0027178302407264714, -0.0017120130360126495 ], "p_value": 0.38035, "mde": 0.00574154155483705, "mde_rel_pct": 10.136634767676144, "verdict": "no significant win", "system_worked": false }, "mechanism_signature": { "prediction": "per-shell expert count spread <= 1", "baseline_mean_load_cv": 0.11686608614250746, "idea_mean_load_cv": 0.0, "baseline_mean_shell_spread": 258.375, "idea_mean_shell_spread": 0.0, "idea_mean_reassigned_fraction": 0.503828125, "confirmed": true }, "custom_track": { "name": "correlated_token_moe_regression", "file": "shell_moe_bench_track.py", "domain": "moe-routing" } }, "how_to_run": "/home/maxwelhelp/main/bin/python3 stage2_shell_moe.py", "files": [ "shell_moe_bench_track.py", "stage2_shell_moe.py", "bench_report.json", "/home/maxwelhelp/all/math2nn/bench/custom_tracks/correlated_token_moe_regression.py" ], "limitations": "The accepted custom track is small and synthetic; no built-in language-model track exists for MoE routing. Top-2 routing, capacity overflow, wall-clock dispatch overhead, and larger-scale perplexity were not tested. The benchmark uses a greedy quota assignment and 12 training epochs.", "system_verdict": "partial", "practical_verdict": "inconclusive", "mechanism_ok": 1, "system_judged": true }