Tiny Local Recurrence with Adaptive Computation / results.json
Unverified
1{
2 "device": "cuda",
3 "mechanism_checks": {
4 "stability": [
5 {
6 "lambda": -26,
7 "q": -1.08,
8 "predicted": "diverge",
9 "observed": "diverge",
10 "abs_s30": 10.06265688907345
11 },
12 {
13 "lambda": -25,
14 "q": -1.0,
15 "predicted": "neutral",
16 "observed": "neutral",
17 "abs_s30": 1.0
18 },
19 {
20 "lambda": -20,
21 "q": -0.6000000000000001,
22 "predicted": "contract",
23 "observed": "contract",
24 "abs_s30": 2.2107391972073433e-07
25 },
26 {
27 "lambda": -13,
28 "q": -0.040000000000000036,
29 "predicted": "contract",
30 "observed": "contract",
31 "abs_s30": 1.1529215046068777e-42
32 },
33 {
34 "lambda": -12.5,
35 "q": 0.0,
36 "predicted": "contract",
37 "observed": "contract",
38 "abs_s30": 0.0
39 },
40 {
41 "lambda": -10,
42 "q": 0.19999999999999996,
43 "predicted": "contract",
44 "observed": "contract",
45 "abs_s30": 1.0737418239999928e-21
46 },
47 {
48 "lambda": 0,
49 "q": 1.0,
50 "predicted": "neutral",
51 "observed": "neutral",
52 "abs_s30": 1.0
53 },
54 {
55 "lambda": 5,
56 "q": 1.4,
57 "predicted": "diverge",
58 "observed": "diverge",
59 "abs_s30": 24201.43235548455
60 }
61 ],
62 "strict_stability_interval": [
63 -25.0,
64 0.0
65 ],
66 "halting_constant_h": [
67 {
68 "h": 0.05,
69 "formula_T": 20,
70 "observed_formula_T": 20,
71 "act_clipped_T": 20
72 },
73 {
74 "h": 0.1,
75 "formula_T": 10,
76 "observed_formula_T": 10,
77 "act_clipped_T": 10
78 },
79 {
80 "h": 0.2,
81 "formula_T": 5,
82 "observed_formula_T": 5,
83 "act_clipped_T": 5
84 },
85 {
86 "h": 0.4,
87 "formula_T": 3,
88 "observed_formula_T": 3,
89 "act_clipped_T": 3
90 },
91 {
92 "h": 0.7,
93 "formula_T": 2,
94 "observed_formula_T": 2,
95 "act_clipped_T": 2
96 }
97 ],
98 "halting_bias_sweep": [
99 {
100 "b": -3,
101 "h": 0.04742587317756678,
102 "formula_T": 22,
103 "observed_T": 22
104 },
105 {
106 "b": -1,
107 "h": 0.2689414213699951,
108 "formula_T": 4,
109 "observed_T": 4
110 },
111 {
112 "b": 0,
113 "h": 0.5,
114 "formula_T": 2,
115 "observed_T": 2
116 },
117 {
118 "b": 1,
119 "h": 0.7310585786300049,
120 "formula_T": 2,
121 "observed_T": 2
122 },
123 {
124 "b": 3,
125 "h": 0.9525741268224334,
126 "formula_T": 2,
127 "observed_T": 2
128 }
129 ]
130 },
131 "mini_experiment": {
132 "baseline": {
133 "accuracy": 0.990234375,
134 "avg_microsteps": 6.0,
135 "train_seconds": 5.6903587950073415,
136 "parameters": 13218
137 },
138 "idea": {
139 "accuracy": 0.990234375,
140 "avg_microsteps": 4.107421875,
141 "train_seconds": 14.9282974020025,
142 "parameters": 3411
143 }
144 }
145}