Hugging Face's logo

Spaces:
open-r1
/
open-r1-eval-leaderboard
Running

App Files Community
1
kashif HF staff commited on
Commit
21f69f0
·
verified ·
1 Parent(s): 96cc850

Upload eval_results/AI-MO/deepseek-math-7b-expo-0.5/main/aimo_kaggle_hard_pot/results_2024-05-28T10-51-26.755380.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. eval_results/AI-MO/deepseek-math-7b-expo-0.5/main/aimo_kaggle_hard_pot/results_2024-05-28T10-51-26.755380.json +187 -0
eval_results/AI-MO/deepseek-math-7b-expo-0.5/main/aimo_kaggle_hard_pot/results_2024-05-28T10-51-26.755380.json ADDED
@@ -0,0 +1,187 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "config_general": {
3
+ "lighteval_sha": "?",
4
+ "num_fewshot_seeds": 1,
5
+ "override_batch_size": 4,
6
+ "max_samples": null,
7
+ "job_id": "",
8
+ "start_time": 53977.944610033,
9
+ "end_time": 54567.736698283,
10
+ "total_evaluation_time_secondes": "589.7920882499966",
11
+ "model_name": "AI-MO/deepseek-math-7b-expo-0.5",
12
+ "model_sha": "7171dc1ad2c73b8f29048819c9d8cf84029c2c74",
13
+ "model_dtype": "torch.bfloat16",
14
+ "model_size": "12.93 GB",
15
+ "config": null
16
+ },
17
+ "results": {
18
+ "custom|aimo_kaggle_hard_pot:v0|0": {
19
+ "qem": 0.02,
20
+ "qem_stderr": 0.02
21
+ },
22
+ "custom|aimo_kaggle_hard_pot:v1|0": {
23
+ "qem": 0.02,
24
+ "qem_stderr": 0.02
25
+ },
26
+ "custom|aimo_kaggle_hard_pot:v2|0": {
27
+ "qem": 0.06,
28
+ "qem_stderr": 0.033926691677251195
29
+ },
30
+ "custom|aimo_kaggle_hard_pot:_average|0": {
31
+ "qem": 0.03333333333333333,
32
+ "qem_stderr": 0.02464223055908373
33
+ },
34
+ "all": {
35
+ "qem": 0.03333333333333333,
36
+ "qem_stderr": 0.02464223055908373
37
+ }
38
+ },
39
+ "versions": {
40
+ "custom|aimo_kaggle_hard_pot:v0|0": 0,
41
+ "custom|aimo_kaggle_hard_pot:v1|0": 0,
42
+ "custom|aimo_kaggle_hard_pot:v2|0": 0
43
+ },
44
+ "config_tasks": {
45
+ "custom|aimo_kaggle_hard_pot:v0": {
46
+ "name": "aimo_kaggle_hard_pot:v0",
47
+ "prompt_function": "kaggle_hard_pot_prompt_fn_v0",
48
+ "hf_repo": "AI-MO/kaggle-validation-set-hard",
49
+ "hf_subset": "v0",
50
+ "metric": [
51
+ "quasi_exact_match_code_and_math"
52
+ ],
53
+ "hf_avail_splits": [
54
+ "train"
55
+ ],
56
+ "evaluation_splits": [
57
+ "train"
58
+ ],
59
+ "few_shots_split": null,
60
+ "few_shots_select": null,
61
+ "generation_size": 2048,
62
+ "stop_sequence": null,
63
+ "output_regex": null,
64
+ "frozen": false,
65
+ "suite": [
66
+ "custom"
67
+ ],
68
+ "original_num_docs": 50,
69
+ "effective_num_docs": 50,
70
+ "trust_dataset": null,
71
+ "must_remove_duplicate_docs": null
72
+ },
73
+ "custom|aimo_kaggle_hard_pot:v1": {
74
+ "name": "aimo_kaggle_hard_pot:v1",
75
+ "prompt_function": "kaggle_hard_pot_prompt_fn_v1",
76
+ "hf_repo": "AI-MO/kaggle-validation-set-hard",
77
+ "hf_subset": "v0",
78
+ "metric": [
79
+ "quasi_exact_match_code_and_math"
80
+ ],
81
+ "hf_avail_splits": [
82
+ "train"
83
+ ],
84
+ "evaluation_splits": [
85
+ "train"
86
+ ],
87
+ "few_shots_split": null,
88
+ "few_shots_select": null,
89
+ "generation_size": 2048,
90
+ "stop_sequence": null,
91
+ "output_regex": null,
92
+ "frozen": false,
93
+ "suite": [
94
+ "custom"
95
+ ],
96
+ "original_num_docs": 50,
97
+ "effective_num_docs": 50,
98
+ "trust_dataset": null,
99
+ "must_remove_duplicate_docs": null
100
+ },
101
+ "custom|aimo_kaggle_hard_pot:v2": {
102
+ "name": "aimo_kaggle_hard_pot:v2",
103
+ "prompt_function": "kaggle_hard_pot_prompt_fn_v2",
104
+ "hf_repo": "AI-MO/kaggle-validation-set-hard",
105
+ "hf_subset": "v0",
106
+ "metric": [
107
+ "quasi_exact_match_code_and_math"
108
+ ],
109
+ "hf_avail_splits": [
110
+ "train"
111
+ ],
112
+ "evaluation_splits": [
113
+ "train"
114
+ ],
115
+ "few_shots_split": null,
116
+ "few_shots_select": null,
117
+ "generation_size": 2048,
118
+ "stop_sequence": null,
119
+ "output_regex": null,
120
+ "frozen": false,
121
+ "suite": [
122
+ "custom"
123
+ ],
124
+ "original_num_docs": 50,
125
+ "effective_num_docs": 50,
126
+ "trust_dataset": null,
127
+ "must_remove_duplicate_docs": null
128
+ }
129
+ },
130
+ "summary_tasks": {
131
+ "custom|aimo_kaggle_hard_pot:v0|0": {
132
+ "hashes": {
133
+ "hash_examples": "303213a38d9f7512",
134
+ "hash_full_prompts": "3c670fa0e80bc301",
135
+ "hash_input_tokens": "1bd4187ad6963415",
136
+ "hash_cont_tokens": "62f9b42d10bc351b"
137
+ },
138
+ "truncated": 50,
139
+ "non_truncated": 0,
140
+ "padded": 37,
141
+ "non_padded": 13,
142
+ "effective_few_shots": 0.0,
143
+ "num_truncated_few_shots": 0
144
+ },
145
+ "custom|aimo_kaggle_hard_pot:v1|0": {
146
+ "hashes": {
147
+ "hash_examples": "e4234b97ad92862f",
148
+ "hash_full_prompts": "09e7759a96f64e59",
149
+ "hash_input_tokens": "dc7f68cf14be3d61",
150
+ "hash_cont_tokens": "c58f48c76f4801e1"
151
+ },
152
+ "truncated": 50,
153
+ "non_truncated": 0,
154
+ "padded": 32,
155
+ "non_padded": 18,
156
+ "effective_few_shots": 0.0,
157
+ "num_truncated_few_shots": 0
158
+ },
159
+ "custom|aimo_kaggle_hard_pot:v2|0": {
160
+ "hashes": {
161
+ "hash_examples": "6396eb8833e13ba0",
162
+ "hash_full_prompts": "65f8b95ea99c7087",
163
+ "hash_input_tokens": "bddb67459e7601c5",
164
+ "hash_cont_tokens": "479740294bb9eac7"
165
+ },
166
+ "truncated": 50,
167
+ "non_truncated": 0,
168
+ "padded": 30,
169
+ "non_padded": 20,
170
+ "effective_few_shots": 0.0,
171
+ "num_truncated_few_shots": 0
172
+ }
173
+ },
174
+ "summary_general": {
175
+ "hashes": {
176
+ "hash_examples": "648c9a107d279e1e",
177
+ "hash_full_prompts": "7f4c38eb08b3bb41",
178
+ "hash_input_tokens": "16f0fd76e0a2a165",
179
+ "hash_cont_tokens": "8a7c490332ca3e61"
180
+ },
181
+ "truncated": 150,
182
+ "non_truncated": 0,
183
+ "padded": 99,
184
+ "non_padded": 51,
185
+ "num_truncated_few_shots": 0
186
+ }
187
+ }