{ "bench_version": 1, "track": "congestion_moe_sequence", "model": "tiny_moe", "metric_direction": "lower is better", "n_seeds": 8, "baseline": { "best_cfg": { "lr": 0.006, "temperature": 0.5 }, "sweep": [ { "cfg": { "lr": 0.0015, "temperature": 0.5 }, "mean": 1.013328194618225 }, { "cfg": { "lr": 0.0015, "temperature": 0.7 }, "mean": 1.0138949006795883 }, { "cfg": { "lr": 0.0015, "temperature": 1.0 }, "mean": 1.0139495730400085 }, { "cfg": { "lr": 0.003, "temperature": 0.5 }, "mean": 0.9528659582138062 }, { "cfg": { "lr": 0.003, "temperature": 0.7 }, "mean": 0.9548223316669464 }, { "cfg": { "lr": 0.003, "temperature": 1.0 }, "mean": 0.9556002467870712 }, { "cfg": { "lr": 0.006, "temperature": 0.5 }, "mean": 0.853488102555275 }, { "cfg": { "lr": 0.006, "temperature": 0.7 }, "mean": 0.859209343791008 }, { "cfg": { "lr": 0.006, "temperature": 1.0 }, "mean": 0.8618226200342178 } ], "full": { "mean": 0.9469610154628754, "std": 0.22496182254054511, "per_seed": [ 0.7249739170074463, 0.5741610527038574, 0.8283010125160217, 1.2865164279937744, 1.0606132745742798, 1.210940957069397, 0.897042989730835, 0.9931384921073914 ], "n": 8 } }, "idea": { "mean": 0.9704978540539742, "std": 0.21903613092802948, "per_seed": [ 0.7384425401687622, 0.5904383659362793, 0.8622050285339355, 1.2669295072555542, 1.1580994129180908, 1.1912760734558105, 0.947786271572113, 1.0088056325912476 ], "n": 8 }, "comparison": { "delta_mean": 0.023536838591098785, "idea_wins": 2, "n_pairs": 8, "per_seed_diffs": [ 0.013468623161315918, 0.016277313232421875, 0.03390401601791382, -0.019586920738220215, 0.09748613834381104, -0.019664883613586426, 0.050743281841278076, 0.0156671404838562 ], "p_value": 0.13795, "mde": 0.03200907134547522, "mde_rel_pct": 3.380188922542831, "verdict": "no significant win", "system_worked": false }, "mechanism_signature": { "idea_sweep": [ { "cfg": { "lr": 0.0015, "temperature": 0.7 }, "result": { "mean": 1.1258931756019592, "std": 0.2915624175884252, "per_seed": [ 0.8899675607681274, 0.7289124131202698, 0.9036810994148254, 1.533188819885254, 1.2896629571914673, 1.5818136930465698, 1.0076074600219727, 1.0723114013671875 ], "n": 8 } }, { "cfg": { "lr": 0.003, "temperature": 0.7 }, "result": { "mean": 1.0657112374901772, "std": 0.26514025298024, "per_seed": [ 0.8271378874778748, 0.6704053282737732, 0.886384129524231, 1.442152738571167, 1.2345929145812988, 1.4378575086593628, 0.9755327105522156, 1.0516266822814941 ], "n": 8 } }, { "cfg": { "lr": 0.006, "temperature": 0.7 }, "result": { "mean": 0.9704978540539742, "std": 0.21903613092802948, "per_seed": [ 0.7384425401687622, 0.5904383659362793, 0.8622050285339355, 1.2669295072555542, 1.1580994129180908, 1.1912760734558105, 0.947786271572113, 1.0088056325912476 ], "n": 8 } } ], "prediction": "congestion-aware routing reduces expert-load CV while retaining a fixed row budget", "predicted_direction": "negative", "observed_load_cv_delta": -0.28653995075335703, "per_seed": [ { "seed": 0, "baseline_load_cv": 0.2926468500132308, "idea_load_cv": 0.0714858382016512, "baseline_mse": 0.7249739170074463, "idea_mse": 0.7384425401687622 }, { "seed": 1, "baseline_load_cv": 0.521491851785781, "idea_load_cv": 0.09097290992372925, "baseline_mse": 0.5741610527038574, "idea_mse": 0.5904383659362793 }, { "seed": 2, "baseline_load_cv": 0.4599689101988864, "idea_load_cv": 0.08442225455900104, "baseline_mse": 0.8283010125160217, "idea_mse": 0.8622050285339355 }, { "seed": 3, "baseline_load_cv": 0.25446022032673565, "idea_load_cv": 0.05582119941488141, "baseline_mse": 1.2865164279937744, "idea_mse": 1.2669295072555542 }, { "seed": 4, "baseline_load_cv": 0.49192661345027733, "idea_load_cv": 0.08442110705389756, "baseline_mse": 1.0606132745742798, "idea_mse": 1.1580994129180908 }, { "seed": 5, "baseline_load_cv": 0.21399866103269982, "idea_load_cv": 0.0416189082410498, "baseline_mse": 1.210940957069397, "idea_mse": 1.1912760734558105 }, { "seed": 6, "baseline_load_cv": 0.16870912551205047, "idea_load_cv": 0.04076074600056684, "baseline_mse": 0.897042989730835, "idea_mse": 0.947786271572113 }, { "seed": 7, "baseline_load_cv": 0.43846935270462917, "idea_load_cv": 0.07984901560265745, "baseline_mse": 0.9931384921073914, "idea_mse": 1.0088056325912476 } ], "confirmed": true }, "custom_track": { "name": "congestion_moe_sequence", "file": "custom_moe_track.py", "domain": "moe-routing" }, "device": "cpu", "epochs": 4, "batch": 128 }