ThomasTheMaker commited on
Commit
2613791
·
verified ·
1 Parent(s): 98c67d3

Upload results_2025-05-10T20-12-49.001669.json with huggingface_hub

Browse files
results_2025-05-10T20-12-49.001669.json CHANGED
@@ -1,33 +1,9 @@
1
  {
2
  "config_general": {
3
- "lighteval_sha": "?",
4
- "num_fewshot_seeds": 1,
5
- "max_samples": null,
6
- "job_id": 0,
7
- "start_time": 619076.153469025,
8
- "end_time": 619462.411040425,
9
  "total_evaluation_time_secondes": "386.2575713999104",
10
  "model_name": "gpt2",
11
  "model_sha": "607a30d783dfa663caf39e06633721c8d4cfcd7e",
12
- "model_dtype": null,
13
- "model_size": "476.2 MB",
14
- "generation_parameters": {
15
- "early_stopping": null,
16
- "repetition_penalty": null,
17
- "frequency_penalty": null,
18
- "length_penalty": null,
19
- "presence_penalty": null,
20
- "max_new_tokens": null,
21
- "min_new_tokens": null,
22
- "seed": null,
23
- "stop_tokens": null,
24
- "temperature": null,
25
- "top_k": null,
26
- "min_p": null,
27
- "top_p": null,
28
- "truncate_prompt": null,
29
- "response_format": null
30
- }
31
  },
32
  "results": {
33
  "leaderboard|truthfulqa:mc|0": {
@@ -45,85 +21,5 @@
45
  },
46
  "versions": {
47
  "leaderboard|truthfulqa:mc|0": 0
48
- },
49
- "config_tasks": {
50
- "leaderboard|truthfulqa:mc": {
51
- "name": "truthfulqa:mc",
52
- "prompt_function": "truthful_qa_multiple_choice",
53
- "hf_repo": "truthful_qa",
54
- "hf_subset": "multiple_choice",
55
- "metric": [
56
- {
57
- "metric_name": [
58
- "truthfulqa_mc1",
59
- "truthfulqa_mc2"
60
- ],
61
- "higher_is_better": {
62
- "truthfulqa_mc1": true,
63
- "truthfulqa_mc2": true
64
- },
65
- "category": "8",
66
- "use_case": "1",
67
- "sample_level_fn": "truthfulqa_mc_metrics",
68
- "corpus_level_fn": {
69
- "truthfulqa_mc1": "mean",
70
- "truthfulqa_mc2": "mean"
71
- }
72
- }
73
- ],
74
- "hf_revision": null,
75
- "hf_filter": null,
76
- "hf_avail_splits": [
77
- "validation"
78
- ],
79
- "trust_dataset": true,
80
- "evaluation_splits": [
81
- "validation"
82
- ],
83
- "few_shots_split": null,
84
- "few_shots_select": null,
85
- "generation_size": -1,
86
- "generation_grammar": null,
87
- "stop_sequence": [
88
- "\n"
89
- ],
90
- "num_samples": null,
91
- "suite": [
92
- "leaderboard"
93
- ],
94
- "original_num_docs": 817,
95
- "effective_num_docs": 817,
96
- "must_remove_duplicate_docs": false,
97
- "version": 0
98
- }
99
- },
100
- "summary_tasks": {
101
- "leaderboard|truthfulqa:mc|0": {
102
- "hashes": {
103
- "hash_examples": "36a6d90e75d92d4a",
104
- "hash_full_prompts": "36a6d90e75d92d4a",
105
- "hash_input_tokens": "1f53251acd4ba722",
106
- "hash_cont_tokens": "e142c03ef8b64bd5"
107
- },
108
- "truncated": 0,
109
- "non_truncated": 817,
110
- "padded": 9619,
111
- "non_padded": 377,
112
- "effective_few_shots": 0.0,
113
- "num_truncated_few_shots": 0
114
- }
115
- },
116
- "summary_general": {
117
- "hashes": {
118
- "hash_examples": "aed1dfc67e53d0f2",
119
- "hash_full_prompts": "aed1dfc67e53d0f2",
120
- "hash_input_tokens": "0a41b966ca38e4d0",
121
- "hash_cont_tokens": "3be708d2914ec601"
122
- },
123
- "truncated": 0,
124
- "non_truncated": 817,
125
- "padded": 9619,
126
- "non_padded": 377,
127
- "num_truncated_few_shots": 0
128
  }
129
  }
 
1
  {
2
  "config_general": {
 
 
 
 
 
 
3
  "total_evaluation_time_secondes": "386.2575713999104",
4
  "model_name": "gpt2",
5
  "model_sha": "607a30d783dfa663caf39e06633721c8d4cfcd7e",
6
+ "model_size": "476.2 MB"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7
  },
8
  "results": {
9
  "leaderboard|truthfulqa:mc|0": {
 
21
  },
22
  "versions": {
23
  "leaderboard|truthfulqa:mc|0": 0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
24
  }
25
  }