Pipelined bounded-staleness gradient coding / bench_report.json
Mechanism confirmed, baseline not beaten
1{
2 "bench_version": 1,
3 "track": "tabular",
4 "model": "mlp_tiny",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.003
10 },
11 "sweep": [
12 {
13 "cfg": {
14 "lr": 0.003
15 },
16 "mean": 8.571802377700806
17 },
18 {
19 "cfg": {
20 "lr": 0.01
21 },
22 "mean": 20.563005447387695
23 },
24 {
25 "cfg": {
26 "lr": 0.03
27 },
28 "mean": 20.28245759010315
29 }
30 ],
31 "full": {
32 "mean": 8.942538976669312,
33 "std": 1.2011059504420323,
34 "per_seed": [
35 8.109341621398926,
36 9.679099082946777,
37 8.425871849060059,
38 8.072896957397461,
39 6.844796180725098,
40 9.526346206665039,
41 10.597946166992188,
42 10.284013748168945
43 ],
44 "n": 8
45 }
46 },
47 "idea": {
48 "mean": 8.995229661464691,
49 "std": 1.2122424602688344,
50 "per_seed": [
51 8.160211563110352,
52 9.738099098205566,
53 8.477083206176758,
54 8.129450798034668,
55 6.85977029800415,
56 9.593474388122559,
57 10.62648868560791,
58 10.377259254455566
59 ],
60 "n": 8
61 },
62 "comparison": {
63 "delta_mean": 0.05269068479537964,
64 "idea_wins": 0,
65 "n_pairs": 8,
66 "per_seed_diffs": [
67 0.05086994171142578,
68 0.05900001525878906,
69 0.05121135711669922,
70 0.05655384063720703,
71 0.014974117279052734,
72 0.06712818145751953,
73 0.028542518615722656,
74 0.0932455062866211
75 ],
76 "p_value": 0.0081,
77 "mde": 0.019784595139020266,
78 "mde_rel_pct": 0.2212413632262314,
79 "verdict": "idea worse (significant)",
80 "system_worked": false
81 },
82 "mechanism_signature": {
83 "predicted": 0.06620380758814955,
84 "observed": 0.008275476713068709,
85 "confirmed": true,
86 "details": {
87 "predicted": 0.06620380758814955,
88 "observed": 0.008275476713068709,
89 "observed_over_predicted": 0.12500001282535644,
90 "mean_gradient_cosine": 0.9849408324128792,
91 "mean_displacement_sq": 1.4717475092518306e-06,
92 "mean_age": 0.4951171875,
93 "max_age": 1,
94 "confirmed": true,
95 "definition": "trained tabular MLP: stale-vs-current partition gradient MSE compared with local L_hat^2 parameter displacement"
96 }
97 },
98 "idea_trials": [
99 {
100 "cfg": {
101 "lr": 0.003,
102 "c": 2
103 },
104 "result": {
105 "mean": 8.995229661464691,
106 "std": 1.2122424602688344,
107 "per_seed": [
108 8.160211563110352,
109 9.738099098205566,
110 8.477083206176758,
111 8.129450798034668,
112 6.85977029800415,
113 9.593474388122559,
114 10.62648868560791,
115 10.377259254455566
116 ],
117 "n": 8
118 }
119 },
120 {
121 "cfg": {
122 "lr": 0.003,
123 "c": 2
124 },
125 "result": {
126 "mean": 8.995229661464691,
127 "std": 1.2122424602688344,
128 "per_seed": [
129 8.160211563110352,
130 9.738099098205566,
131 8.477083206176758,
132 8.129450798034668,
133 6.85977029800415,
134 9.593474388122559,
135 10.62648868560791,
136 10.377259254455566
137 ],
138 "n": 8
139 }
140 },
141 {
142 "cfg": {
143 "lr": 0.03,
144 "c": 2
145 },
146 "result": {
147 "mean": 24.871363401412964,
148 "std": 9.356500961042444,
149 "per_seed": [
150 27.6580810546875,
151 45.721439361572266,
152 10.339054107666016,
153 22.65554428100586,
154 22.052181243896484,
155 25.39988136291504,
156 19.183732986450195,
157 25.96099281311035
158 ],
159 "n": 8
160 }
161 }
162 ],
163 "custom_track": null
164}