Hugging Face's logo

minhleduc
/
rlcd-e2-checkpoints

Safetensors
Model card Files
xet
 Community
minhleduc commited on
Commit
3631704
·
verified ·
1 Parent(s): cfbd908

e2_ce_only_seed44: test_eval_v2.json

Browse files
Files changed (1) hide show
  1. e2_ce_only_seed44/test_eval_v2.json +99 -0
e2_ce_only_seed44/test_eval_v2.json ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "raw": {
3
+ "ece": 0.13831696626544,
4
+ "brier": 0.051284364823037706,
5
+ "nll": 0.8615161191206425,
6
+ "accuracy": 0.783,
7
+ "soft_accuracy": 0.5222538067623973,
8
+ "mean_confidence": 0.6452848957926035,
9
+ "mean_target_max": 0.6589462240487337,
10
+ "score_mae": 0.21527907053823583,
11
+ "n": 2000,
12
+ "by_type": {
13
+ "choice": {
14
+ "ece": 0.15338748335838318,
15
+ "brier": 0.06259250025831231,
16
+ "nll": 0.986831498829027,
17
+ "accuracy": 0.765,
18
+ "soft_accuracy": 0.47661630041897296,
19
+ "mean_confidence": 0.6136187381048998,
20
+ "mean_target_max": 0.6345381752649943,
21
+ "score_mae": null,
22
+ "n": 600
23
+ },
24
+ "score": {
25
+ "ece": 0.1602796625718474,
26
+ "brier": 0.05328848799827483,
27
+ "nll": 1.0365782652422786,
28
+ "accuracy": 0.74375,
29
+ "soft_accuracy": 0.44316409012302754,
30
+ "mean_confidence": 0.5840223825722933,
31
+ "mean_target_max": 0.5928848401084542,
32
+ "score_mae": 0.21527907053823583,
33
+ "n": 800
34
+ },
35
+ "noul": {
36
+ "ece": 0.09865541060765583,
37
+ "brier": 0.0373040651541136,
38
+ "nll": 0.50278454458341,
39
+ "accuracy": 0.8533333333333334,
40
+ "soft_accuracy": 0.6733442686249812,
41
+ "mean_confidence": 0.7586344044407208,
42
+ "mean_target_max": 0.7714361180861791,
43
+ "score_mae": null,
44
+ "n": 600
45
+ }
46
+ }
47
+ },
48
+ "post_temperature": {
49
+ "ece": 0.1613641293048859,
50
+ "brier": 0.04948239230941744,
51
+ "nll": 0.8574540826445446,
52
+ "accuracy": 0.783,
53
+ "soft_accuracy": 0.5113731906339526,
54
+ "mean_confidence": 0.6228788203001022,
55
+ "mean_target_max": 0.6589462240487337,
56
+ "score_mae": 0.21538325707893818,
57
+ "n": 2000,
58
+ "by_type": {
59
+ "choice": {
60
+ "ece": 0.19152938291430474,
61
+ "brier": 0.06042450366864311,
62
+ "nll": 0.9820380164931218,
63
+ "accuracy": 0.765,
64
+ "soft_accuracy": 0.45812685097257294,
65
+ "mean_confidence": 0.5765153206884861,
66
+ "mean_target_max": 0.6345381752649943,
67
+ "score_mae": null,
68
+ "n": 600
69
+ },
70
+ "score": {
71
+ "ece": 0.17983196027576925,
72
+ "brier": 0.05104208130190152,
73
+ "nll": 1.031392307500355,
74
+ "accuracy": 0.74375,
75
+ "soft_accuracy": 0.43448590984568,
76
+ "mean_confidence": 0.564741894416511,
77
+ "mean_target_max": 0.5928848401084542,
78
+ "score_mae": 0.21538325707893818,
79
+ "n": 800
80
+ },
81
+ "noul": {
82
+ "ece": 0.11253386884927749,
83
+ "brier": 0.03646069562687969,
84
+ "nll": 0.5009525156548867,
85
+ "accuracy": 0.8533333333333334,
86
+ "soft_accuracy": 0.6671359046796957,
87
+ "mean_confidence": 0.7467582210898399,
88
+ "mean_target_max": 0.7714361180861791,
89
+ "score_mae": null,
90
+ "n": 600
91
+ }
92
+ }
93
+ },
94
+ "fitted_temperature": {
95
+ "type1_bucket1": 1.0892750024795532,
96
+ "type0_bucket1": 1.1707441806793213,
97
+ "type2_bucket0": 1.0939185619354248
98
+ }
99
+ }