Hugging Face's logo

minhleduc
/
rlcd-e2-checkpoints

Safetensors
Model card Files
xet
 Community
minhleduc commited on
Commit
3d9da7a
·
verified ·
1 Parent(s): cdcabf1

e2_laya_rlce_seed42: test_eval_v2.json

Browse files
Files changed (1) hide show
  1. e2_laya_rlce_seed42/test_eval_v2.json +99 -0
e2_laya_rlce_seed42/test_eval_v2.json ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "raw": {
3
+ "ece": 0.12944456613063812,
4
+ "brier": 0.052770321191238,
5
+ "nll": 0.8640999112045392,
6
+ "accuracy": 0.7725,
7
+ "soft_accuracy": 0.5200601447559893,
8
+ "mean_confidence": 0.6430554198771715,
9
+ "mean_target_max": 0.6589462240487337,
10
+ "score_mae": 0.22544640515581704,
11
+ "n": 2000,
12
+ "by_type": {
13
+ "choice": {
14
+ "ece": 0.13134420956174533,
15
+ "brier": 0.06734055001695462,
16
+ "nll": 0.996281831363837,
17
+ "accuracy": 0.7383333333333333,
18
+ "soft_accuracy": 0.476577321117123,
19
+ "mean_confidence": 0.619853642731905,
20
+ "mean_target_max": 0.6345381752649943,
21
+ "score_mae": null,
22
+ "n": 600
23
+ },
24
+ "score": {
25
+ "ece": 0.15436594255268574,
26
+ "brier": 0.05415011412187596,
27
+ "nll": 1.0367637212667615,
28
+ "accuracy": 0.72625,
29
+ "soft_accuracy": 0.4360020062606782,
30
+ "mean_confidence": 0.5718840314447879,
31
+ "mean_target_max": 0.5928848401084542,
32
+ "score_mae": 0.22544640515581704,
33
+ "n": 800
34
+ },
35
+ "noul": {
36
+ "ece": 0.10718097815910976,
37
+ "brier": 0.03636036845800408,
38
+ "nll": 0.5016995776289453,
39
+ "accuracy": 0.8683333333333333,
40
+ "soft_accuracy": 0.6756204863886038,
41
+ "mean_confidence": 0.7611523815989494,
42
+ "mean_target_max": 0.7714361180861791,
43
+ "score_mae": null,
44
+ "n": 600
45
+ }
46
+ }
47
+ },
48
+ "post_temperature": {
49
+ "ece": 0.14125536119937895,
50
+ "brier": 0.05157535602751921,
51
+ "nll": 0.8613823981322348,
52
+ "accuracy": 0.7725,
53
+ "soft_accuracy": 0.5142514189705253,
54
+ "mean_confidence": 0.6312446529269219,
55
+ "mean_target_max": 0.6589462240487337,
56
+ "score_mae": 0.2255610384931788,
57
+ "n": 2000,
58
+ "by_type": {
59
+ "choice": {
60
+ "ece": 0.1449350022772948,
61
+ "brier": 0.06476511814263025,
62
+ "nll": 0.9905648699216545,
63
+ "accuracy": 0.7383333333333333,
64
+ "soft_accuracy": 0.4659412851681312,
65
+ "mean_confidence": 0.5977527806162835,
66
+ "mean_target_max": 0.6345381752649943,
67
+ "score_mae": null,
68
+ "n": 600
69
+ },
70
+ "score": {
71
+ "ece": 0.1571685499697924,
72
+ "brier": 0.053852836025589566,
73
+ "nll": 1.0360501168854535,
74
+ "accuracy": 0.72625,
75
+ "soft_accuracy": 0.43474282953888177,
76
+ "mean_confidence": 0.5690814781188965,
77
+ "mean_target_max": 0.5928848401084542,
78
+ "score_mae": 0.2255610384931788,
79
+ "n": 800
80
+ },
81
+ "noul": {
82
+ "ece": 0.12071254024902979,
83
+ "brier": 0.03534895391498104,
84
+ "nll": 0.49930963467185696,
85
+ "accuracy": 0.8683333333333333,
86
+ "soft_accuracy": 0.668573005348444,
87
+ "mean_confidence": 0.7476207583149275,
88
+ "mean_target_max": 0.7714361180861791,
89
+ "score_mae": null,
90
+ "n": 600
91
+ }
92
+ }
93
+ },
94
+ "fitted_temperature": {
95
+ "type0_bucket1": 1.0957692861557007,
96
+ "type2_bucket0": 1.1091790199279785,
97
+ "type1_bucket1": 1.0124728679656982
98
+ }
99
+ }