Hugging Face's logo

minhleduc
/
rlcd-e2-checkpoints

Safetensors
Model card Files
xet
 Community
minhleduc commited on
Commit
0295eaa
·
verified ·
1 Parent(s): 04a4d20

e2_ce_only_seed42: test_eval_v2.json

Browse files
Files changed (1) hide show
  1. e2_ce_only_seed42/test_eval_v2.json +99 -0
e2_ce_only_seed42/test_eval_v2.json ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "raw": {
3
+ "ece": 0.1398480086773634,
4
+ "brier": 0.05103822000877925,
5
+ "nll": 0.8604203677009791,
6
+ "accuracy": 0.7855,
7
+ "soft_accuracy": 0.5235573233515024,
8
+ "mean_confidence": 0.6458810180872678,
9
+ "mean_target_max": 0.6589462240487337,
10
+ "score_mae": 0.21578076076169964,
11
+ "n": 2000,
12
+ "by_type": {
13
+ "choice": {
14
+ "ece": 0.14743219847480457,
15
+ "brier": 0.06339079040756132,
16
+ "nll": 0.9857434277236462,
17
+ "accuracy": 0.7633333333333333,
18
+ "soft_accuracy": 0.47836515128612517,
19
+ "mean_confidence": 0.6159011376897494,
20
+ "mean_target_max": 0.6345381752649943,
21
+ "score_mae": null,
22
+ "n": 600
23
+ },
24
+ "score": {
25
+ "ece": 0.16289205316454172,
26
+ "brier": 0.05298609249110598,
27
+ "nll": 1.0357897561136633,
28
+ "accuracy": 0.74625,
29
+ "soft_accuracy": 0.4441511295735836,
30
+ "mean_confidence": 0.584728074260056,
31
+ "mean_target_max": 0.5928848401084542,
32
+ "score_mae": 0.21578076076169964,
33
+ "n": 800
34
+ },
35
+ "noul": {
36
+ "ece": 0.10260182797908782,
37
+ "brier": 0.036088486300228206,
38
+ "nll": 0.5012714564613998,
39
+ "accuracy": 0.86,
40
+ "soft_accuracy": 0.6746244204541048,
41
+ "mean_confidence": 0.7573981569210688,
42
+ "mean_target_max": 0.7714361180861791,
43
+ "score_mae": null,
44
+ "n": 600
45
+ }
46
+ }
47
+ },
48
+ "post_temperature": {
49
+ "ece": 0.1588473957479,
50
+ "brier": 0.04917404281355277,
51
+ "nll": 0.8560326781123877,
52
+ "accuracy": 0.7855,
53
+ "soft_accuracy": 0.5143366709724069,
54
+ "mean_confidence": 0.6268481774032116,
55
+ "mean_target_max": 0.6589462240487337,
56
+ "score_mae": 0.2155858551512938,
57
+ "n": 2000,
58
+ "by_type": {
59
+ "choice": {
60
+ "ece": 0.1697476014494896,
61
+ "brier": 0.06110521121062883,
62
+ "nll": 0.9806674488882223,
63
+ "accuracy": 0.7633333333333333,
64
+ "soft_accuracy": 0.4674839984625578,
65
+ "mean_confidence": 0.5935857443511486,
66
+ "mean_target_max": 0.6345381752649943,
67
+ "score_mae": null,
68
+ "n": 600
69
+ },
70
+ "score": {
71
+ "ece": 0.1827498009428382,
72
+ "brier": 0.05077533899652735,
73
+ "nll": 1.0302575784828514,
74
+ "accuracy": 0.74625,
75
+ "soft_accuracy": 0.4355850401893258,
76
+ "mean_confidence": 0.5657574540749192,
77
+ "mean_target_max": 0.5928848401084542,
78
+ "score_mae": 0.2155858551512938,
79
+ "n": 800
80
+ },
81
+ "noul": {
82
+ "ece": 0.11843508313099545,
83
+ "brier": 0.035107812839177244,
84
+ "nll": 0.4990980401759346,
85
+ "accuracy": 0.86,
86
+ "soft_accuracy": 0.6661915178596973,
87
+ "mean_confidence": 0.741564908226331,
88
+ "mean_target_max": 0.7714361180861791,
89
+ "score_mae": null,
90
+ "n": 600
91
+ }
92
+ }
93
+ },
94
+ "fitted_temperature": {
95
+ "type0_bucket1": 1.1001685857772827,
96
+ "type2_bucket0": 1.130467176437378,
97
+ "type1_bucket1": 1.0884613990783691
98
+ }
99
+ }