Upload eval_results/deepseek-ai/deepseek-math-7b-rl/main/aimo_kaggle_hard_pot/results_2024-05-28T09-06-30.789083.json with huggingface_hub
Browse files
eval_results/deepseek-ai/deepseek-math-7b-rl/main/aimo_kaggle_hard_pot/results_2024-05-28T09-06-30.789083.json
ADDED
@@ -0,0 +1,193 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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": 2154508.660483309,
|
9 |
+
"end_time": 2155107.785841121,
|
10 |
+
"total_evaluation_time_secondes": "599.1253578118049",
|
11 |
+
"model_name": "deepseek-ai/deepseek-math-7b-rl",
|
12 |
+
"model_sha": "f3cd419a172ff5a100e9a563ebfe7900d78e9740",
|
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": 0.04,
|
28 |
+
"qem_stderr": 0.02799416848895062
|
29 |
+
},
|
30 |
+
"custom|aimo_kaggle_hard_pot:_average|0": {
|
31 |
+
"qem": 0.02666666666666667,
|
32 |
+
"qem_stderr": 0.022664722829650208
|
33 |
+
},
|
34 |
+
"all": {
|
35 |
+
"qem": 0.02666666666666667,
|
36 |
+
"qem_stderr": 0.022664722829650208
|
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 |
+
"num_samples": null,
|
65 |
+
"frozen": false,
|
66 |
+
"suite": [
|
67 |
+
"custom"
|
68 |
+
],
|
69 |
+
"original_num_docs": 50,
|
70 |
+
"effective_num_docs": 50,
|
71 |
+
"trust_dataset": null,
|
72 |
+
"must_remove_duplicate_docs": null,
|
73 |
+
"version": 0
|
74 |
+
},
|
75 |
+
"custom|aimo_kaggle_hard_pot:v1": {
|
76 |
+
"name": "aimo_kaggle_hard_pot:v1",
|
77 |
+
"prompt_function": "kaggle_hard_pot_prompt_fn_v1",
|
78 |
+
"hf_repo": "AI-MO/kaggle-validation-set-hard",
|
79 |
+
"hf_subset": "v0",
|
80 |
+
"metric": [
|
81 |
+
"quasi_exact_match_code_and_math"
|
82 |
+
],
|
83 |
+
"hf_avail_splits": [
|
84 |
+
"train"
|
85 |
+
],
|
86 |
+
"evaluation_splits": [
|
87 |
+
"train"
|
88 |
+
],
|
89 |
+
"few_shots_split": null,
|
90 |
+
"few_shots_select": null,
|
91 |
+
"generation_size": 2048,
|
92 |
+
"stop_sequence": null,
|
93 |
+
"output_regex": null,
|
94 |
+
"num_samples": null,
|
95 |
+
"frozen": false,
|
96 |
+
"suite": [
|
97 |
+
"custom"
|
98 |
+
],
|
99 |
+
"original_num_docs": 50,
|
100 |
+
"effective_num_docs": 50,
|
101 |
+
"trust_dataset": null,
|
102 |
+
"must_remove_duplicate_docs": null,
|
103 |
+
"version": 0
|
104 |
+
},
|
105 |
+
"custom|aimo_kaggle_hard_pot:v2": {
|
106 |
+
"name": "aimo_kaggle_hard_pot:v2",
|
107 |
+
"prompt_function": "kaggle_hard_pot_prompt_fn_v2",
|
108 |
+
"hf_repo": "AI-MO/kaggle-validation-set-hard",
|
109 |
+
"hf_subset": "v0",
|
110 |
+
"metric": [
|
111 |
+
"quasi_exact_match_code_and_math"
|
112 |
+
],
|
113 |
+
"hf_avail_splits": [
|
114 |
+
"train"
|
115 |
+
],
|
116 |
+
"evaluation_splits": [
|
117 |
+
"train"
|
118 |
+
],
|
119 |
+
"few_shots_split": null,
|
120 |
+
"few_shots_select": null,
|
121 |
+
"generation_size": 2048,
|
122 |
+
"stop_sequence": null,
|
123 |
+
"output_regex": null,
|
124 |
+
"num_samples": null,
|
125 |
+
"frozen": false,
|
126 |
+
"suite": [
|
127 |
+
"custom"
|
128 |
+
],
|
129 |
+
"original_num_docs": 50,
|
130 |
+
"effective_num_docs": 50,
|
131 |
+
"trust_dataset": null,
|
132 |
+
"must_remove_duplicate_docs": null,
|
133 |
+
"version": 0
|
134 |
+
}
|
135 |
+
},
|
136 |
+
"summary_tasks": {
|
137 |
+
"custom|aimo_kaggle_hard_pot:v0|0": {
|
138 |
+
"hashes": {
|
139 |
+
"hash_examples": "303213a38d9f7512",
|
140 |
+
"hash_full_prompts": "3c670fa0e80bc301",
|
141 |
+
"hash_input_tokens": "1bd4187ad6963415",
|
142 |
+
"hash_cont_tokens": "4a5131314ac71945"
|
143 |
+
},
|
144 |
+
"truncated": 50,
|
145 |
+
"non_truncated": 0,
|
146 |
+
"padded": 37,
|
147 |
+
"non_padded": 13,
|
148 |
+
"effective_few_shots": 0.0,
|
149 |
+
"num_truncated_few_shots": 0
|
150 |
+
},
|
151 |
+
"custom|aimo_kaggle_hard_pot:v1|0": {
|
152 |
+
"hashes": {
|
153 |
+
"hash_examples": "e4234b97ad92862f",
|
154 |
+
"hash_full_prompts": "09e7759a96f64e59",
|
155 |
+
"hash_input_tokens": "dc7f68cf14be3d61",
|
156 |
+
"hash_cont_tokens": "835e43e6085c79ec"
|
157 |
+
},
|
158 |
+
"truncated": 50,
|
159 |
+
"non_truncated": 0,
|
160 |
+
"padded": 32,
|
161 |
+
"non_padded": 18,
|
162 |
+
"effective_few_shots": 0.0,
|
163 |
+
"num_truncated_few_shots": 0
|
164 |
+
},
|
165 |
+
"custom|aimo_kaggle_hard_pot:v2|0": {
|
166 |
+
"hashes": {
|
167 |
+
"hash_examples": "6396eb8833e13ba0",
|
168 |
+
"hash_full_prompts": "65f8b95ea99c7087",
|
169 |
+
"hash_input_tokens": "bddb67459e7601c5",
|
170 |
+
"hash_cont_tokens": "16c7b73b8b022a8d"
|
171 |
+
},
|
172 |
+
"truncated": 50,
|
173 |
+
"non_truncated": 0,
|
174 |
+
"padded": 30,
|
175 |
+
"non_padded": 20,
|
176 |
+
"effective_few_shots": 0.0,
|
177 |
+
"num_truncated_few_shots": 0
|
178 |
+
}
|
179 |
+
},
|
180 |
+
"summary_general": {
|
181 |
+
"hashes": {
|
182 |
+
"hash_examples": "648c9a107d279e1e",
|
183 |
+
"hash_full_prompts": "7f4c38eb08b3bb41",
|
184 |
+
"hash_input_tokens": "16f0fd76e0a2a165",
|
185 |
+
"hash_cont_tokens": "c21eba382df462d7"
|
186 |
+
},
|
187 |
+
"truncated": 150,
|
188 |
+
"non_truncated": 0,
|
189 |
+
"padded": 99,
|
190 |
+
"non_padded": 51,
|
191 |
+
"num_truncated_few_shots": 0
|
192 |
+
}
|
193 |
+
}
|