Spaces:
Building
Building
Upload eval_results/AI-MO/deepseek-math-7b-rl-sft/aimo_v00.12/mini_math_v2_mmos/results_2024-05-17T06-08-10.872266.json with huggingface_hub
Browse files
eval_results/AI-MO/deepseek-math-7b-rl-sft/aimo_v00.12/mini_math_v2_mmos/results_2024-05-17T06-08-10.872266.json
ADDED
@@ -0,0 +1,286 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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": 114051.201789419,
|
9 |
+
"end_time": 114961.813979436,
|
10 |
+
"total_evaluation_time_secondes": "910.6121900169965",
|
11 |
+
"model_name": "AI-MO/deepseek-math-7b-rl-sft",
|
12 |
+
"model_sha": "3c503cc5c1046c6751f6b92c21aef5b8dfc96673",
|
13 |
+
"model_dtype": "torch.bfloat16",
|
14 |
+
"model_size": "12.93 GB",
|
15 |
+
"config": null
|
16 |
+
},
|
17 |
+
"results": {
|
18 |
+
"custom|mini_math_v2_mmos:level_1|0": {
|
19 |
+
"qem": 0.6,
|
20 |
+
"qem_stderr": 0.08401680504168059
|
21 |
+
},
|
22 |
+
"custom|mini_math_v2_mmos:level_2|0": {
|
23 |
+
"qem": 0.4583333333333333,
|
24 |
+
"qem_stderr": 0.05913268547421811
|
25 |
+
},
|
26 |
+
"custom|mini_math_v2_mmos:level_3|0": {
|
27 |
+
"qem": 0.3888888888888889,
|
28 |
+
"qem_stderr": 0.05167468693203862
|
29 |
+
},
|
30 |
+
"custom|mini_math_v2_mmos:level_4|0": {
|
31 |
+
"qem": 0.29896907216494845,
|
32 |
+
"qem_stderr": 0.046724655123323536
|
33 |
+
},
|
34 |
+
"custom|mini_math_v2_mmos:level_5|0": {
|
35 |
+
"qem": 0.19811320754716982,
|
36 |
+
"qem_stderr": 0.03889722288318549
|
37 |
+
},
|
38 |
+
"custom|mini_math_v2_mmos:_average|0": {
|
39 |
+
"qem": 0.3888609003868681,
|
40 |
+
"qem_stderr": 0.05608921109088927
|
41 |
+
},
|
42 |
+
"all": {
|
43 |
+
"qem": 0.3888609003868681,
|
44 |
+
"qem_stderr": 0.05608921109088927
|
45 |
+
}
|
46 |
+
},
|
47 |
+
"versions": {
|
48 |
+
"custom|mini_math_v2_mmos:level_1|0": 0,
|
49 |
+
"custom|mini_math_v2_mmos:level_2|0": 0,
|
50 |
+
"custom|mini_math_v2_mmos:level_3|0": 0,
|
51 |
+
"custom|mini_math_v2_mmos:level_4|0": 0,
|
52 |
+
"custom|mini_math_v2_mmos:level_5|0": 0
|
53 |
+
},
|
54 |
+
"config_tasks": {
|
55 |
+
"custom|mini_math_v2_mmos:level_1": {
|
56 |
+
"name": "mini_math_v2_mmos:level_1",
|
57 |
+
"prompt_function": "minimath_mmos_prompt_fn",
|
58 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
59 |
+
"hf_subset": "Level 1",
|
60 |
+
"metric": [
|
61 |
+
"quasi_exact_match_code_and_math"
|
62 |
+
],
|
63 |
+
"hf_avail_splits": [
|
64 |
+
"train",
|
65 |
+
"test"
|
66 |
+
],
|
67 |
+
"evaluation_splits": [
|
68 |
+
"test"
|
69 |
+
],
|
70 |
+
"few_shots_split": null,
|
71 |
+
"few_shots_select": null,
|
72 |
+
"generation_size": 2048,
|
73 |
+
"stop_sequence": null,
|
74 |
+
"output_regex": null,
|
75 |
+
"frozen": false,
|
76 |
+
"suite": [
|
77 |
+
"custom"
|
78 |
+
],
|
79 |
+
"original_num_docs": 35,
|
80 |
+
"effective_num_docs": 35,
|
81 |
+
"trust_dataset": null,
|
82 |
+
"must_remove_duplicate_docs": null
|
83 |
+
},
|
84 |
+
"custom|mini_math_v2_mmos:level_2": {
|
85 |
+
"name": "mini_math_v2_mmos:level_2",
|
86 |
+
"prompt_function": "minimath_mmos_prompt_fn",
|
87 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
88 |
+
"hf_subset": "Level 2",
|
89 |
+
"metric": [
|
90 |
+
"quasi_exact_match_code_and_math"
|
91 |
+
],
|
92 |
+
"hf_avail_splits": [
|
93 |
+
"train",
|
94 |
+
"test"
|
95 |
+
],
|
96 |
+
"evaluation_splits": [
|
97 |
+
"test"
|
98 |
+
],
|
99 |
+
"few_shots_split": null,
|
100 |
+
"few_shots_select": null,
|
101 |
+
"generation_size": 2048,
|
102 |
+
"stop_sequence": null,
|
103 |
+
"output_regex": null,
|
104 |
+
"frozen": false,
|
105 |
+
"suite": [
|
106 |
+
"custom"
|
107 |
+
],
|
108 |
+
"original_num_docs": 72,
|
109 |
+
"effective_num_docs": 72,
|
110 |
+
"trust_dataset": null,
|
111 |
+
"must_remove_duplicate_docs": null
|
112 |
+
},
|
113 |
+
"custom|mini_math_v2_mmos:level_3": {
|
114 |
+
"name": "mini_math_v2_mmos:level_3",
|
115 |
+
"prompt_function": "minimath_mmos_prompt_fn",
|
116 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
117 |
+
"hf_subset": "Level 3",
|
118 |
+
"metric": [
|
119 |
+
"quasi_exact_match_code_and_math"
|
120 |
+
],
|
121 |
+
"hf_avail_splits": [
|
122 |
+
"train",
|
123 |
+
"test"
|
124 |
+
],
|
125 |
+
"evaluation_splits": [
|
126 |
+
"test"
|
127 |
+
],
|
128 |
+
"few_shots_split": null,
|
129 |
+
"few_shots_select": null,
|
130 |
+
"generation_size": 2048,
|
131 |
+
"stop_sequence": null,
|
132 |
+
"output_regex": null,
|
133 |
+
"frozen": false,
|
134 |
+
"suite": [
|
135 |
+
"custom"
|
136 |
+
],
|
137 |
+
"original_num_docs": 90,
|
138 |
+
"effective_num_docs": 90,
|
139 |
+
"trust_dataset": null,
|
140 |
+
"must_remove_duplicate_docs": null
|
141 |
+
},
|
142 |
+
"custom|mini_math_v2_mmos:level_4": {
|
143 |
+
"name": "mini_math_v2_mmos:level_4",
|
144 |
+
"prompt_function": "minimath_mmos_prompt_fn",
|
145 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
146 |
+
"hf_subset": "Level 4",
|
147 |
+
"metric": [
|
148 |
+
"quasi_exact_match_code_and_math"
|
149 |
+
],
|
150 |
+
"hf_avail_splits": [
|
151 |
+
"train",
|
152 |
+
"test"
|
153 |
+
],
|
154 |
+
"evaluation_splits": [
|
155 |
+
"test"
|
156 |
+
],
|
157 |
+
"few_shots_split": null,
|
158 |
+
"few_shots_select": null,
|
159 |
+
"generation_size": 2048,
|
160 |
+
"stop_sequence": null,
|
161 |
+
"output_regex": null,
|
162 |
+
"frozen": false,
|
163 |
+
"suite": [
|
164 |
+
"custom"
|
165 |
+
],
|
166 |
+
"original_num_docs": 97,
|
167 |
+
"effective_num_docs": 97,
|
168 |
+
"trust_dataset": null,
|
169 |
+
"must_remove_duplicate_docs": null
|
170 |
+
},
|
171 |
+
"custom|mini_math_v2_mmos:level_5": {
|
172 |
+
"name": "mini_math_v2_mmos:level_5",
|
173 |
+
"prompt_function": "minimath_mmos_prompt_fn",
|
174 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
175 |
+
"hf_subset": "Level 5",
|
176 |
+
"metric": [
|
177 |
+
"quasi_exact_match_code_and_math"
|
178 |
+
],
|
179 |
+
"hf_avail_splits": [
|
180 |
+
"train",
|
181 |
+
"test"
|
182 |
+
],
|
183 |
+
"evaluation_splits": [
|
184 |
+
"test"
|
185 |
+
],
|
186 |
+
"few_shots_split": null,
|
187 |
+
"few_shots_select": null,
|
188 |
+
"generation_size": 2048,
|
189 |
+
"stop_sequence": null,
|
190 |
+
"output_regex": null,
|
191 |
+
"frozen": false,
|
192 |
+
"suite": [
|
193 |
+
"custom"
|
194 |
+
],
|
195 |
+
"original_num_docs": 106,
|
196 |
+
"effective_num_docs": 106,
|
197 |
+
"trust_dataset": null,
|
198 |
+
"must_remove_duplicate_docs": null
|
199 |
+
}
|
200 |
+
},
|
201 |
+
"summary_tasks": {
|
202 |
+
"custom|mini_math_v2_mmos:level_1|0": {
|
203 |
+
"hashes": {
|
204 |
+
"hash_examples": "ad7e0d89fb7b0664",
|
205 |
+
"hash_full_prompts": "09389803d825fdf3",
|
206 |
+
"hash_input_tokens": "417bc9ee4fcab798",
|
207 |
+
"hash_cont_tokens": "9aee3ccdfbd10354"
|
208 |
+
},
|
209 |
+
"truncated": 35,
|
210 |
+
"non_truncated": 0,
|
211 |
+
"padded": 24,
|
212 |
+
"non_padded": 11,
|
213 |
+
"effective_few_shots": 0.0,
|
214 |
+
"num_truncated_few_shots": 0
|
215 |
+
},
|
216 |
+
"custom|mini_math_v2_mmos:level_2|0": {
|
217 |
+
"hashes": {
|
218 |
+
"hash_examples": "493b7e3130e3a50d",
|
219 |
+
"hash_full_prompts": "f5f9ef766a19cf57",
|
220 |
+
"hash_input_tokens": "3fafd8c7323f3bb4",
|
221 |
+
"hash_cont_tokens": "54a882aede824485"
|
222 |
+
},
|
223 |
+
"truncated": 72,
|
224 |
+
"non_truncated": 0,
|
225 |
+
"padded": 37,
|
226 |
+
"non_padded": 35,
|
227 |
+
"effective_few_shots": 0.0,
|
228 |
+
"num_truncated_few_shots": 0
|
229 |
+
},
|
230 |
+
"custom|mini_math_v2_mmos:level_3|0": {
|
231 |
+
"hashes": {
|
232 |
+
"hash_examples": "11edfc7fc00756f4",
|
233 |
+
"hash_full_prompts": "716fbe1386202336",
|
234 |
+
"hash_input_tokens": "f5c3171f680a2e9a",
|
235 |
+
"hash_cont_tokens": "ef08dafd86c575a5"
|
236 |
+
},
|
237 |
+
"truncated": 90,
|
238 |
+
"non_truncated": 0,
|
239 |
+
"padded": 37,
|
240 |
+
"non_padded": 53,
|
241 |
+
"effective_few_shots": 0.0,
|
242 |
+
"num_truncated_few_shots": 0
|
243 |
+
},
|
244 |
+
"custom|mini_math_v2_mmos:level_4|0": {
|
245 |
+
"hashes": {
|
246 |
+
"hash_examples": "a901e1669616ae50",
|
247 |
+
"hash_full_prompts": "32cb1558a580e38f",
|
248 |
+
"hash_input_tokens": "6d5f32ab97f76f8b",
|
249 |
+
"hash_cont_tokens": "5b7cebf7bdf536ef"
|
250 |
+
},
|
251 |
+
"truncated": 96,
|
252 |
+
"non_truncated": 1,
|
253 |
+
"padded": 34,
|
254 |
+
"non_padded": 63,
|
255 |
+
"effective_few_shots": 0.0,
|
256 |
+
"num_truncated_few_shots": 0
|
257 |
+
},
|
258 |
+
"custom|mini_math_v2_mmos:level_5|0": {
|
259 |
+
"hashes": {
|
260 |
+
"hash_examples": "1d024f2e2410736e",
|
261 |
+
"hash_full_prompts": "4b1345f7301730ff",
|
262 |
+
"hash_input_tokens": "58a467e662a3f415",
|
263 |
+
"hash_cont_tokens": "27b97a77ceac5602"
|
264 |
+
},
|
265 |
+
"truncated": 105,
|
266 |
+
"non_truncated": 1,
|
267 |
+
"padded": 27,
|
268 |
+
"non_padded": 79,
|
269 |
+
"effective_few_shots": 0.0,
|
270 |
+
"num_truncated_few_shots": 0
|
271 |
+
}
|
272 |
+
},
|
273 |
+
"summary_general": {
|
274 |
+
"hashes": {
|
275 |
+
"hash_examples": "06e45ee0bae45a44",
|
276 |
+
"hash_full_prompts": "7c1d25a76def20ed",
|
277 |
+
"hash_input_tokens": "7e79c0c0764c2c82",
|
278 |
+
"hash_cont_tokens": "573e42e22e695085"
|
279 |
+
},
|
280 |
+
"truncated": 398,
|
281 |
+
"non_truncated": 2,
|
282 |
+
"padded": 159,
|
283 |
+
"non_padded": 241,
|
284 |
+
"num_truncated_few_shots": 0
|
285 |
+
}
|
286 |
+
}
|