Mode-Aware Mask Schedule / bench_report.json
Failed on benchmark
1{
2 "bench_version": 1,
3 "track": "masked_multitoken_modes",
4 "model": "transformer_tiny",
5 "metric_direction": "lower is better",
6 "n_seeds": 8,
7 "baseline": {
8 "best_cfg": {
9 "lr": 0.002,
10 "epochs": 14,
11 "hide_count": 1
12 },
13 "sweep": [
14 {
15 "cfg": {
16 "lr": 0.002,
17 "epochs": 14,
18 "hide_count": 1
19 },
20 "mean": 0.1428266018629074
21 },
22 {
23 "cfg": {
24 "lr": 0.004,
25 "epochs": 14,
26 "hide_count": 1
27 },
28 "mean": 0.20178561471402645
29 },
30 {
31 "cfg": {
32 "lr": 0.008,
33 "epochs": 14,
34 "hide_count": 1
35 },
36 "mean": 0.14775850623846054
37 }
38 ],
39 "full": {
40 "mean": 0.14373092539608479,
41 "std": 0.02385596096955299,
42 "per_seed": [
43 0.1670890897512436,
44 0.12829364836215973,
45 0.13317705690860748,
46 0.14274661242961884,
47 0.16149330139160156,
48 0.09596626460552216,
49 0.14481675624847412,
50 0.1762646734714508
51 ],
52 "n": 8
53 }
54 },
55 "idea": {
56 "mean": 0.14993123710155487,
57 "std": 0.038141062047303144,
58 "per_seed": [
59 0.1759413331747055,
60 0.13947612047195435,
61 0.12913769483566284,
62 0.14234721660614014,
63 0.149277925491333,
64 0.09132890403270721,
65 0.13944129645824432,
66 0.2324994057416916
67 ],
68 "n": 8,
69 "best_cfg": {
70 "lr": 0.004,
71 "epochs": 14,
72 "lam": 0.1,
73 "rho": 0.01
74 }
75 },
76 "comparison": {
77 "delta_mean": 0.006200311705470085,
78 "idea_wins": 5,
79 "n_pairs": 8,
80 "per_seed_diffs": [
81 0.008852243423461914,
82 0.011182472109794617,
83 -0.004039362072944641,
84 -0.00039939582347869873,
85 -0.012215375900268555,
86 -0.004637360572814941,
87 -0.005375459790229797,
88 0.056234732270240784
89 ],
90 "p_value": 0.63155,
91 "mde": 0.018084067093522574,
92 "mde_rel_pct": 12.581890114244809,
93 "verdict": "no significant win",
94 "system_worked": false
95 },
96 "mechanism_signature": {
97 "schedule_check": [
98 {
99 "lambda": 0,
100 "rho": 0,
101 "pred_pi1": 0,
102 "obs_pi1": 0.0,
103 "pred_full": 0,
104 "obs_full": 0.0,
105 "v_min": 7
106 },
107 {
108 "lambda": 0.1,
109 "rho": 0,
110 "pred_pi1": 0.1,
111 "obs_pi1": 0.10140000283718109,
112 "pred_full": 0,
113 "obs_full": 0.0,
114 "v_min": 0
115 },
116 {
117 "lambda": 0.1,
118 "rho": 0.01,
119 "pred_pi1": 0.11,
120 "obs_pi1": 0.11155000329017639,
121 "pred_full": 0.01,
122 "obs_full": 0.009879999794065952,
123 "v_min": 0
124 },
125 {
126 "lambda": 0.25,
127 "rho": 0.01,
128 "pred_pi1": 0.26,
129 "obs_pi1": 0.26243001222610474,
130 "pred_full": 0.01,
131 "obs_full": 0.009879999794065952,
132 "v_min": 0
133 }
134 ],
135 "signature": {
136 "quantity": "empty-context mode probability",
137 "target_true_weight": 0.8,
138 "predicted_baseline": 0.8209748044610023,
139 "predicted_idea": 0.839116208255291,
140 "observed_baseline": 0.8209748044610023,
141 "observed_idea": 0.839116208255291,
142 "predicted_improvement": -0.018141403794288635,
143 "confirmed": false
144 },
145 "custom_track": {
146 "name": "masked_multitoken_modes",
147 "file": "custom_masked_track.py",
148 "domain": "sequence"
149 }
150 },
151 "idea_sweep": [
152 {
153 "cfg": {
154 "lr": 0.002,
155 "epochs": 14,
156 "lam": 0.1,
157 "rho": 0.0
158 },
159 "mean": 0.16454514022916555,
160 "per_seed": [
161 0.16334068775177002,
162 0.22535602748394012,
163 0.12713226675987244,
164 0.18480685353279114,
165 0.19457480311393738,
166 0.10892865806818008,
167 0.14529073238372803,
168 0.16693109273910522
169 ],
170 "mode_probs": [
171 0.7183287739753723,
172 0.8767662644386292,
173 0.9214068055152893,
174 0.8836508393287659,
175 0.8848533034324646,
176 0.7462123036384583,
177 0.780626654624939,
178 0.8711745738983154
179 ]
180 },
181 {
182 "cfg": {
183 "lr": 0.004,
184 "epochs": 14,
185 "lam": 0.1,
186 "rho": 0.01
187 },
188 "mean": 0.14993123710155487,
189 "per_seed": [
190 0.1759413331747055,
191 0.13947612047195435,
192 0.12913769483566284,
193 0.14234721660614014,
194 0.149277925491333,
195 0.09132890403270721,
196 0.13944129645824432,
197 0.2324994057416916
198 ],
199 "mode_probs": [
200 0.9315788745880127,
201 0.8918474316596985,
202 0.8515940308570862,
203 0.9011403918266296,
204 0.7730483412742615,
205 0.7809845209121704,
206 0.7969619035720825,
207 0.7857741713523865
208 ]
209 },
210 {
211 "cfg": {
212 "lr": 0.008,
213 "epochs": 14,
214 "lam": 0.25,
215 "rho": 0.01
216 },
217 "mean": 0.18514209240674973,
218 "per_seed": [
219 0.18203185498714447,
220 0.16977907717227936,
221 0.13093167543411255,
222 0.2541387677192688,
223 0.21239057183265686,
224 0.16110031306743622,
225 0.14696472883224487,
226 0.22379975020885468
227 ],
228 "mode_probs": [
229 0.7211427092552185,
230 0.8639533519744873,
231 0.8087149262428284,
232 0.8416880965232849,
233 0.8448388576507568,
234 0.8393517136573792,
235 0.7988701462745667,
236 0.8555705547332764
237 ]
238 }
239 ]
240}