Confidence-Tested LoRA Pruning / results.json
Failed on benchmark
1{
2 "mechanism": {
3 "prediction_1_se_slope": {
4 "predicted": -0.5,
5 "observed": -0.4958754521864339,
6 "se_by_T": [
7 0.09849592864076846,
8 0.07063183797122778,
9 0.049859882043533324,
10 0.0352658929861761,
11 0.02499620997363385
12 ]
13 },
14 "prediction_2_hac_inflation": {
15 "rho": [
16 0.0,
17 0.3,
18 0.6,
19 0.8
20 ],
21 "predicted": [
22 1.0,
23 1.8571428571428574,
24 4.0,
25 9.000000000000002
26 ],
27 "observed": [
28 1.0,
29 2.143685032163134,
30 4.654398731246324,
31 9.346017526509169
32 ]
33 },
34 "prediction_3_rejection_power": {
35 "true_mean": [
36 0.45,
37 0.3,
38 0.2,
39 0.05
40 ],
41 "delta": 0.25,
42 "rejection_rate": [
43 0.0,
44 0.022222222222222223,
45 0.17777777777777778,
46 0.7111111111111111
47 ]
48 }
49 },
50 "pruning": {
51 "keep": 10,
52 "delta": 0.2,
53 "repeats": 30,
54 "statistical_indices": [
55 12,
56 13,
57 14,
58 15,
59 16,
60 18,
61 19,
62 21,
63 22,
64 23
65 ],
66 "latest_indices": [
67 0,
68 1,
69 2,
70 4,
71 8,
72 10,
73 15,
74 16,
75 17,
76 21
77 ],
78 "statistical_true_utility_mean": 3.6241159420289852,
79 "latest_true_utility_mean": 2.9755942028985505,
80 "statistical_win_rate": 0.9333333333333333,
81 "all_true_utility": 3.939130434782608,
82 "statistical_pvalues": [
83 0.07103683114410965,
84 0.056811674304565116,
85 0.28195511385390604,
86 0.12142901774514786,
87 0.09593979057710517,
88 0.7627441479190542,
89 0.16687679537699113,
90 0.27328870275768863,
91 0.2095243957387451,
92 0.5092997790761873,
93 0.14418556209625216,
94 0.8337167257204463,
95 0.9533451388260267,
96 0.9785266826003848,
97 0.9587336611538738,
98 0.9897164477819104,
99 0.988714473008437,
100 0.8896582003792419,
101 0.9615325154263084,
102 0.9227973866458504,
103 0.9136451445166478,
104 0.9860504633316789,
105 0.9804159594288162,
106 0.9977862894458377
107 ]
108 }
109}