Agnostic Geometry-Prior Mixer / results.json
Beats tuned baseline
1{
2 "device": "cuda",
3 "math_check": {
4 "convex_interpolation_inside_bounds": true,
5 "d_mean_alpha_d_beta_at_zero": 0.25,
6 "initial_alpha_beta_minus4": 0.01798621006309986,
7 "suppression_gradient_sign": true
8 },
9 "results": {
10 "correct": {
11 "free": {
12 "rmse": 0.13216720521450043,
13 "free_branch_rmse": 0.13216720521450043,
14 "geom_branch_rmse": 0.13216720521450043,
15 "mean_alpha": 0.0,
16 "gate_entropy": 1.7310190969510586e-06
17 },
18 "geom": {
19 "rmse": 0.011932441964745522,
20 "free_branch_rmse": 0.011932441964745522,
21 "geom_branch_rmse": 0.011932441964745522,
22 "mean_alpha": 1.0,
23 "gate_entropy": 1.7310190969510586e-06
24 },
25 "concat": {
26 "rmse": 0.01182486955076456,
27 "free_branch_rmse": 0.01182486955076456,
28 "geom_branch_rmse": 0.01182486955076456,
29 "mean_alpha": 1.0,
30 "gate_entropy": 1.7310190969510586e-06
31 },
32 "global": {
33 "rmse": 0.029833368957042694,
34 "free_branch_rmse": 0.3897355794906616,
35 "geom_branch_rmse": 10.78345775604248,
36 "mean_alpha": 0.03491225093603134,
37 "gate_entropy": 0.15142330527305603
38 },
39 "mixer": {
40 "rmse": 0.010266127064824104,
41 "free_branch_rmse": 0.41977766156196594,
42 "geom_branch_rmse": 0.27537474036216736,
43 "mean_alpha": 0.6074919700622559,
44 "gate_entropy": 0.6697317361831665
45 },
46 "global_lambda_0.02": {
47 "rmse": 0.021190768107771873,
48 "free_branch_rmse": 0.42418360710144043,
49 "geom_branch_rmse": 12.315999984741211,
50 "mean_alpha": 0.03323910012841225,
51 "gate_entropy": 0.14582732319831848
52 }
53 },
54 "wrong": {
55 "free": {
56 "rmse": 0.06733770668506622,
57 "free_branch_rmse": 0.06733770668506622,
58 "geom_branch_rmse": 0.06733770668506622,
59 "mean_alpha": 0.0,
60 "gate_entropy": 1.7310190969510586e-06
61 },
62 "geom": {
63 "rmse": 0.06290043145418167,
64 "free_branch_rmse": 0.06290043145418167,
65 "geom_branch_rmse": 0.06290043145418167,
66 "mean_alpha": 1.0,
67 "gate_entropy": 1.7310190969510586e-06
68 },
69 "concat": {
70 "rmse": 0.07601242512464523,
71 "free_branch_rmse": 0.07601242512464523,
72 "geom_branch_rmse": 0.07601242512464523,
73 "mean_alpha": 1.0,
74 "gate_entropy": 1.7310190969510586e-06
75 },
76 "global": {
77 "rmse": 0.0752580538392067,
78 "free_branch_rmse": 0.08121613413095474,
79 "geom_branch_rmse": 0.6568082571029663,
80 "mean_alpha": 0.044423263520002365,
81 "gate_entropy": 0.18175531923770905
82 },
83 "mixer": {
84 "rmse": 0.029876839369535446,
85 "free_branch_rmse": 0.4139192998409271,
86 "geom_branch_rmse": 1.8147188425064087,
87 "mean_alpha": 0.1801556646823883,
88 "gate_entropy": 0.41717299818992615
89 },
90 "global_lambda_0.02": {
91 "rmse": 0.03819429129362106,
92 "free_branch_rmse": 0.046958278864622116,
93 "geom_branch_rmse": 1.1284319162368774,
94 "mean_alpha": 0.02048567123711109,
95 "gate_entropy": 0.09992331266403198
96 }
97 },
98 "none": {
99 "free": {
100 "rmse": 0.06042604148387909,
101 "free_branch_rmse": 0.06042604148387909,
102 "geom_branch_rmse": 0.06042604148387909,
103 "mean_alpha": 0.0,
104 "gate_entropy": 1.7310190969510586e-06
105 },
106 "geom": {
107 "rmse": 0.050840113312006,
108 "free_branch_rmse": 0.050840113312006,
109 "geom_branch_rmse": 0.050840113312006,
110 "mean_alpha": 1.0,
111 "gate_entropy": 1.7310190969510586e-06
112 },
113 "concat": {
114 "rmse": 0.06793663650751114,
115 "free_branch_rmse": 0.06793663650751114,
116 "geom_branch_rmse": 0.06793663650751114,
117 "mean_alpha": 1.0,
118 "gate_entropy": 1.7310190969510586e-06
119 },
120 "global": {
121 "rmse": 0.060362111777067184,
122 "free_branch_rmse": 0.10155397653579712,
123 "geom_branch_rmse": 2.0144524574279785,
124 "mean_alpha": 0.03692173585295677,
125 "gate_entropy": 0.15803472697734833
126 },
127 "mixer": {
128 "rmse": 0.041153136640787125,
129 "free_branch_rmse": 0.14096201956272125,
130 "geom_branch_rmse": 0.7465592622756958,
131 "mean_alpha": 0.10575458407402039,
132 "gate_entropy": 0.32703617215156555
133 },
134 "global_lambda_0.02": {
135 "rmse": 0.08374335616827011,
136 "free_branch_rmse": 0.4030120074748993,
137 "geom_branch_rmse": 9.209486961364746,
138 "mean_alpha": 0.04062730446457863,
139 "gate_entropy": 0.16993264853954315
140 }
141 },
142 "half": {
143 "free": {
144 "rmse": 0.16700385510921478,
145 "free_branch_rmse": 0.16700385510921478,
146 "geom_branch_rmse": 0.16700385510921478,
147 "mean_alpha": 0.0,
148 "gate_entropy": 1.7310190969510586e-06
149 },
150 "geom": {
151 "rmse": 0.12250637263059616,
152 "free_branch_rmse": 0.12250637263059616,
153 "geom_branch_rmse": 0.12250637263059616,
154 "mean_alpha": 1.0,
155 "gate_entropy": 1.7310190969510586e-06
156 },
157 "concat": {
158 "rmse": 0.13126066327095032,
159 "free_branch_rmse": 0.13126066327095032,
160 "geom_branch_rmse": 0.13126066327095032,
161 "mean_alpha": 1.0,
162 "gate_entropy": 1.7310190969510586e-06
163 },
164 "global": {
165 "rmse": 0.15563292801380157,
166 "free_branch_rmse": 0.559171199798584,
167 "geom_branch_rmse": 7.377834796905518,
168 "mean_alpha": 0.06807496398687363,
169 "gate_entropy": 0.24863073229789734
170 },
171 "mixer": {
172 "rmse": 0.12156328558921814,
173 "free_branch_rmse": 1.4644699096679688,
174 "geom_branch_rmse": 0.5877335071563721,
175 "mean_alpha": 0.7179057002067566,
176 "gate_entropy": 0.5880767703056335
177 },
178 "global_lambda_0.02": {
179 "rmse": 0.14033490419387817,
180 "free_branch_rmse": 0.5717989802360535,
181 "geom_branch_rmse": 6.770478248596191,
182 "mean_alpha": 0.07580511271953583,
183 "gate_entropy": 0.2684025168418884
184 }
185 }
186 }
187}