Hugging Face's logo

minhleduc
/
rlcd-e2-checkpoints

Safetensors
Model card Files
xet
 Community
minhleduc commited on
Commit
e4ed5bd
·
verified ·
1 Parent(s): d1512da

e2_rl_only_seed42: test_eval_v2.json

Browse files
Files changed (1) hide show
  1. e2_rl_only_seed42/test_eval_v2.json +99 -0
e2_rl_only_seed42/test_eval_v2.json ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "raw": {
3
+ "ece": 0.12679152297228574,
4
+ "brier": 0.05535062948582825,
5
+ "nll": 0.8679260851321742,
6
+ "accuracy": 0.76,
7
+ "soft_accuracy": 0.5145708695724607,
8
+ "mean_confidence": 0.6345017634481191,
9
+ "mean_target_max": 0.6589462240487337,
10
+ "score_mae": 0.2358442002814263,
11
+ "n": 2000,
12
+ "by_type": {
13
+ "choice": {
14
+ "ece": 0.12754969368378322,
15
+ "brier": 0.06870347295043758,
16
+ "nll": 0.9975430295554301,
17
+ "accuracy": 0.73,
18
+ "soft_accuracy": 0.4690246512244145,
19
+ "mean_confidence": 0.6079408426582813,
20
+ "mean_target_max": 0.6345381752649943,
21
+ "score_mae": null,
22
+ "n": 600
23
+ },
24
+ "score": {
25
+ "ece": 0.15719266310334207,
26
+ "brier": 0.05798107115449511,
27
+ "nll": 1.044197057574056,
28
+ "accuracy": 0.71875,
29
+ "soft_accuracy": 0.43173864085227254,
30
+ "mean_confidence": 0.5640414332598448,
31
+ "mean_target_max": 0.5928848401084542,
32
+ "score_mae": 0.2358442002814263,
33
+ "n": 800
34
+ },
35
+ "noul": {
36
+ "ece": 0.10082647318641345,
37
+ "brier": 0.03849053046299644,
38
+ "nll": 0.5032811774530759,
39
+ "accuracy": 0.845,
40
+ "soft_accuracy": 0.6705600595474244,
41
+ "mean_confidence": 0.7550097911556561,
42
+ "mean_target_max": 0.7714361180861791,
43
+ "score_mae": null,
44
+ "n": 600
45
+ }
46
+ }
47
+ },
48
+ "post_temperature": {
49
+ "ece": 0.14100187775492665,
50
+ "brier": 0.05418899545646501,
51
+ "nll": 0.8656314948331565,
52
+ "accuracy": 0.76,
53
+ "soft_accuracy": 0.5069824938178062,
54
+ "mean_confidence": 0.6189981058388948,
55
+ "mean_target_max": 0.6589462240487337,
56
+ "score_mae": 0.2364325567521155,
57
+ "n": 2000,
58
+ "by_type": {
59
+ "choice": {
60
+ "ece": 0.140854502171278,
61
+ "brier": 0.06703519375523986,
62
+ "nll": 0.9948613907334706,
63
+ "accuracy": 0.73,
64
+ "soft_accuracy": 0.46041965164244175,
65
+ "mean_confidence": 0.5903157998124758,
66
+ "mean_target_max": 0.6345381752649943,
67
+ "score_mae": null,
68
+ "n": 600
69
+ },
70
+ "score": {
71
+ "ece": 0.1665362735465169,
72
+ "brier": 0.05691809966142955,
73
+ "nll": 1.0416682769916952,
74
+ "accuracy": 0.71875,
75
+ "soft_accuracy": 0.4263803570531309,
76
+ "mean_confidence": 0.552213729210198,
77
+ "mean_target_max": 0.5928848401084542,
78
+ "score_mae": 0.2364325567521155,
79
+ "n": 800
80
+ },
81
+ "noul": {
82
+ "ece": 0.12232253074645996,
83
+ "brier": 0.037703991551070765,
84
+ "nll": 0.5016858893881242,
85
+ "accuracy": 0.845,
86
+ "soft_accuracy": 0.6610148516794045,
87
+ "mean_confidence": 0.7367262473702431,
88
+ "mean_target_max": 0.7714361180861791,
89
+ "score_mae": null,
90
+ "n": 600
91
+ }
92
+ }
93
+ },
94
+ "fitted_temperature": {
95
+ "type0_bucket1": 1.0745879411697388,
96
+ "type2_bucket0": 1.1491918563842773,
97
+ "type1_bucket1": 1.0556284189224243
98
+ }
99
+ }