Hugging Face's logo

minhleduc
/
rlcd-e2-checkpoints

Safetensors
Model card Files
xet
 Community
minhleduc commited on
Commit
c143ea2
·
verified ·
1 Parent(s): bd46335

e2_laya_rlce_seed44: test_eval_v2.json

Browse files
Files changed (1) hide show
  1. e2_laya_rlce_seed44/test_eval_v2.json +99 -0
e2_laya_rlce_seed44/test_eval_v2.json ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "raw": {
3
+ "ece": 0.13515584364533428,
4
+ "brier": 0.05498821342930109,
5
+ "nll": 0.8672727304939181,
6
+ "accuracy": 0.7725,
7
+ "soft_accuracy": 0.5162278877273202,
8
+ "mean_confidence": 0.6376278363466262,
9
+ "mean_target_max": 0.6589462240487337,
10
+ "score_mae": 0.22886333604110404,
11
+ "n": 2000,
12
+ "by_type": {
13
+ "choice": {
14
+ "ece": 0.1287393668293953,
15
+ "brier": 0.06880938887503968,
16
+ "nll": 0.9961432211908201,
17
+ "accuracy": 0.745,
18
+ "soft_accuracy": 0.4730249839524428,
19
+ "mean_confidence": 0.6162606234848499,
20
+ "mean_target_max": 0.6345381752649943,
21
+ "score_mae": null,
22
+ "n": 600
23
+ },
24
+ "score": {
25
+ "ece": 0.17092002514749766,
26
+ "brier": 0.057593915954716975,
27
+ "nll": 1.0435723754111677,
28
+ "accuracy": 0.7375,
29
+ "soft_accuracy": 0.4324589177593589,
30
+ "mean_confidence": 0.5673982724919915,
31
+ "mean_target_max": 0.5928848401084542,
32
+ "score_mae": 0.22886333604110404,
33
+ "n": 800
34
+ },
35
+ "noul": {
36
+ "ece": 0.09428338597218196,
37
+ "brier": 0.037692767949674646,
38
+ "nll": 0.5033360465740164,
39
+ "accuracy": 0.8466666666666667,
40
+ "soft_accuracy": 0.6711227514594793,
41
+ "mean_confidence": 0.752634467681249,
42
+ "mean_target_max": 0.7714361180861791,
43
+ "score_mae": null,
44
+ "n": 600
45
+ }
46
+ }
47
+ },
48
+ "post_temperature": {
49
+ "ece": 0.1516125675588846,
50
+ "brier": 0.0535372811920504,
51
+ "nll": 0.8644865690991282,
52
+ "accuracy": 0.7725,
53
+ "soft_accuracy": 0.5081833966597915,
54
+ "mean_confidence": 0.6208874048888683,
55
+ "mean_target_max": 0.6589462240487337,
56
+ "score_mae": 0.2301318886788795,
57
+ "n": 2000,
58
+ "by_type": {
59
+ "choice": {
60
+ "ece": 0.15398936261733376,
61
+ "brier": 0.06616318898743581,
62
+ "nll": 0.9919862444822987,
63
+ "accuracy": 0.745,
64
+ "soft_accuracy": 0.46095144987106323,
65
+ "mean_confidence": 0.5910106347997983,
66
+ "mean_target_max": 0.6345381752649943,
67
+ "score_mae": null,
68
+ "n": 600
69
+ },
70
+ "score": {
71
+ "ece": 0.18403407491743565,
72
+ "brier": 0.05635196869279753,
73
+ "nll": 1.0409193228930236,
74
+ "accuracy": 0.7375,
75
+ "soft_accuracy": 0.42615522114560006,
76
+ "mean_confidence": 0.5534659356623888,
77
+ "mean_target_max": 0.5928848401084542,
78
+ "score_mae": 0.2301318886788795,
79
+ "n": 800
80
+ },
81
+ "noul": {
82
+ "ece": 0.1078599774837494,
83
+ "brier": 0.03715845672900216,
84
+ "nll": 0.5017432219907642,
85
+ "accuracy": 0.8466666666666667,
86
+ "soft_accuracy": 0.6647862441341083,
87
+ "mean_confidence": 0.740659467279911,
88
+ "mean_target_max": 0.7714361180861791,
89
+ "score_mae": null,
90
+ "n": 600
91
+ }
92
+ }
93
+ },
94
+ "fitted_temperature": {
95
+ "type1_bucket1": 1.0644937753677368,
96
+ "type0_bucket1": 1.1086640357971191,
97
+ "type2_bucket0": 1.0983814001083374
98
+ }
99
+ }