PowerMachine commited on
Commit
356c192
·
verified ·
1 Parent(s): 6d1848d

Delete v6_5_7ds_attn_v3_predict_fix_eval.json

Browse files
v6_5_7ds_attn_v3_predict_fix_eval.json DELETED
@@ -1,153 +0,0 @@
1
- {
2
- "evaluation": "predict_fix_v65",
3
- "user_requirement": "rótulos 'gato'/'cachorro' fixos → extração variável e flexível",
4
- "n_test_queries": 6,
5
- "results": [
6
- {
7
- "query": "o gato dorme na cama",
8
- "prediction": "body",
9
- "prediction_proba": "body",
10
- "probability": 0.9416029453277588,
11
- "is_gato_hardcoded": false,
12
- "is_cachorro_hardcoded": false,
13
- "is_registry_label": true,
14
- "is_default_label": false
15
- },
16
- {
17
- "query": "calcule dois mais dois",
18
- "prediction": "body",
19
- "prediction_proba": "body",
20
- "probability": 0.9416029453277588,
21
- "is_gato_hardcoded": false,
22
- "is_cachorro_hardcoded": false,
23
- "is_registry_label": true,
24
- "is_default_label": false
25
- },
26
- {
27
- "query": "qual é a capital do brasil",
28
- "prediction": "body",
29
- "prediction_proba": "body",
30
- "probability": 0.9416029453277588,
31
- "is_gato_hardcoded": false,
32
- "is_cachorro_hardcoded": false,
33
- "is_registry_label": true,
34
- "is_default_label": false
35
- },
36
- {
37
- "query": "explique o que é uma rede neural",
38
- "prediction": "body",
39
- "prediction_proba": "body",
40
- "probability": 0.9416029453277588,
41
- "is_gato_hardcoded": false,
42
- "is_cachorro_hardcoded": false,
43
- "is_registry_label": true,
44
- "is_default_label": false
45
- },
46
- {
47
- "query": "olá como você está",
48
- "prediction": "body",
49
- "prediction_proba": "body",
50
- "probability": 0.9416029453277588,
51
- "is_gato_hardcoded": false,
52
- "is_cachorro_hardcoded": false,
53
- "is_registry_label": true,
54
- "is_default_label": false
55
- },
56
- {
57
- "query": "traduza hello para portugues",
58
- "prediction": "body",
59
- "prediction_proba": "body",
60
- "probability": 0.9416029453277588,
61
- "is_gato_hardcoded": false,
62
- "is_cachorro_hardcoded": false,
63
- "is_registry_label": true,
64
- "is_default_label": false
65
- }
66
- ],
67
- "per_dataset_results": [
68
- {
69
- "dataset": "dominguesm/restore-punctuation-ptbr-dataset",
70
- "expected_labels": [
71
- "unpunctuated",
72
- "punctuated"
73
- ],
74
- "prediction": "punctuated",
75
- "valid": true
76
- },
77
- {
78
- "dataset": "carolina-c4ai/corpus-carolina",
79
- "expected_labels": [
80
- "raw_corpus",
81
- "normalized_text"
82
- ],
83
- "prediction": "normalized_text",
84
- "valid": true
85
- },
86
- {
87
- "dataset": "CEIA-POSITIVO/ultrachat_br_clustred_balanced_v1",
88
- "expected_labels": [
89
- "user_turn",
90
- "assistant_turn"
91
- ],
92
- "prediction": "assistant_turn",
93
- "valid": true
94
- },
95
- {
96
- "dataset": "dominguesm/Canarim-Instruct-PTBR-Dataset",
97
- "expected_labels": [
98
- "instruction",
99
- "response"
100
- ],
101
- "prediction": "response",
102
- "valid": true
103
- },
104
- {
105
- "dataset": "adalbertojunior/punctuation-ptbr",
106
- "expected_labels": [
107
- "unpunctuated",
108
- "punctuated"
109
- ],
110
- "prediction": "punctuated",
111
- "valid": true
112
- },
113
- {
114
- "dataset": "iara-project/news-articles-ptbr-dataset",
115
- "expected_labels": [
116
- "headline",
117
- "body"
118
- ],
119
- "prediction": "body",
120
- "valid": true
121
- },
122
- {
123
- "dataset": "manoela/noticias_ptbr",
124
- "expected_labels": [
125
- "headline",
126
- "body"
127
- ],
128
- "prediction": "body",
129
- "valid": true
130
- }
131
- ],
132
- "summary": {
133
- "n_returns_gato": 0,
134
- "n_returns_cachorro": 0,
135
- "n_returns_registry_label": 6,
136
- "n_returns_default_label": 0,
137
- "fix_pass": true,
138
- "n_datasets_registered": 7,
139
- "last_dataset_used": "manoela/noticias_ptbr"
140
- },
141
- "mathematical_correction": {
142
- "before_bug": "return 'gato' if prob <= 0.5 else 'cachorro' # FIXO",
143
- "after_fix": "label_int = 1 if prob > 0.5 else 0; label_str = self.get_label_string(label_int, dataset_name)",
144
- "registry_lookup": "label_registry[dataset_name][label_int]",
145
- "fallback_chain": [
146
- "1. registry[dataset_name][label_int]",
147
- "2. registry[last_dataset][label_int]",
148
- "3. default {0: 'negative', 1: 'positive'}",
149
- "4. synthetic f'label_{label_int}'"
150
- ]
151
- },
152
- "verdict": "PASS"
153
- }