Upload eval_results/AI-MO/deepseek-math-7b-sft/aimo_v00.00/math/results_2024-04-24T13-57-53.176523.json with huggingface_hub
Browse files
eval_results/AI-MO/deepseek-math-7b-sft/aimo_v00.00/math/results_2024-04-24T13-57-53.176523.json
ADDED
@@ -0,0 +1,410 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"config_general": {
|
3 |
+
"lighteval_sha": "?",
|
4 |
+
"num_fewshot_seeds": 1,
|
5 |
+
"override_batch_size": 2,
|
6 |
+
"max_samples": 10,
|
7 |
+
"job_id": "",
|
8 |
+
"start_time": 3652558.79848515,
|
9 |
+
"end_time": 3652962.355642761,
|
10 |
+
"total_evaluation_time_secondes": "403.55715761100873",
|
11 |
+
"model_name": "AI-MO/deepseek-math-7b-sft",
|
12 |
+
"model_sha": "748a5f5e457052be6c9476d87222e596fedefdb7",
|
13 |
+
"model_dtype": "torch.bfloat16",
|
14 |
+
"model_size": "12.93 GB",
|
15 |
+
"config": null
|
16 |
+
},
|
17 |
+
"results": {
|
18 |
+
"lighteval|math:algebra|0": {
|
19 |
+
"qem": 0.7,
|
20 |
+
"qem_stderr": 0.15275252316519466
|
21 |
+
},
|
22 |
+
"lighteval|math:counting_and_probability|0": {
|
23 |
+
"qem": 0.5,
|
24 |
+
"qem_stderr": 0.16666666666666666
|
25 |
+
},
|
26 |
+
"lighteval|math:geometry|0": {
|
27 |
+
"qem": 0.3,
|
28 |
+
"qem_stderr": 0.15275252316519466
|
29 |
+
},
|
30 |
+
"lighteval|math:intermediate_algebra|0": {
|
31 |
+
"qem": 0.1,
|
32 |
+
"qem_stderr": 0.09999999999999999
|
33 |
+
},
|
34 |
+
"lighteval|math:number_theory|0": {
|
35 |
+
"qem": 0.2,
|
36 |
+
"qem_stderr": 0.13333333333333333
|
37 |
+
},
|
38 |
+
"lighteval|math:prealgebra|0": {
|
39 |
+
"qem": 0.6,
|
40 |
+
"qem_stderr": 0.16329931618554522
|
41 |
+
},
|
42 |
+
"lighteval|math:precalculus|0": {
|
43 |
+
"qem": 0.4,
|
44 |
+
"qem_stderr": 0.1632993161855452
|
45 |
+
},
|
46 |
+
"lighteval|math:_average|0": {
|
47 |
+
"qem": 0.39999999999999997,
|
48 |
+
"qem_stderr": 0.14744338267163995
|
49 |
+
},
|
50 |
+
"all": {
|
51 |
+
"qem": 0.39999999999999997,
|
52 |
+
"qem_stderr": 0.14744338267163995
|
53 |
+
}
|
54 |
+
},
|
55 |
+
"versions": {
|
56 |
+
"lighteval|math:algebra|0": 0,
|
57 |
+
"lighteval|math:counting_and_probability|0": 0,
|
58 |
+
"lighteval|math:geometry|0": 0,
|
59 |
+
"lighteval|math:intermediate_algebra|0": 0,
|
60 |
+
"lighteval|math:number_theory|0": 0,
|
61 |
+
"lighteval|math:prealgebra|0": 0,
|
62 |
+
"lighteval|math:precalculus|0": 0
|
63 |
+
},
|
64 |
+
"config_tasks": {
|
65 |
+
"lighteval|math:algebra": {
|
66 |
+
"name": "math:algebra",
|
67 |
+
"prompt_function": "math",
|
68 |
+
"hf_repo": "lighteval/MATH",
|
69 |
+
"hf_subset": "algebra",
|
70 |
+
"metric": [
|
71 |
+
"quasi_exact_match_math"
|
72 |
+
],
|
73 |
+
"hf_avail_splits": [
|
74 |
+
"train",
|
75 |
+
"test",
|
76 |
+
"validation"
|
77 |
+
],
|
78 |
+
"evaluation_splits": [
|
79 |
+
"test"
|
80 |
+
],
|
81 |
+
"few_shots_split": null,
|
82 |
+
"few_shots_select": null,
|
83 |
+
"generation_size": 2048,
|
84 |
+
"stop_sequence": [
|
85 |
+
"\n"
|
86 |
+
],
|
87 |
+
"output_regex": null,
|
88 |
+
"frozen": false,
|
89 |
+
"suite": [
|
90 |
+
"lighteval",
|
91 |
+
"math"
|
92 |
+
],
|
93 |
+
"original_num_docs": 1187,
|
94 |
+
"effective_num_docs": 10,
|
95 |
+
"trust_dataset": true,
|
96 |
+
"must_remove_duplicate_docs": null
|
97 |
+
},
|
98 |
+
"lighteval|math:counting_and_probability": {
|
99 |
+
"name": "math:counting_and_probability",
|
100 |
+
"prompt_function": "math",
|
101 |
+
"hf_repo": "lighteval/MATH",
|
102 |
+
"hf_subset": "counting_and_probability",
|
103 |
+
"metric": [
|
104 |
+
"quasi_exact_match_math"
|
105 |
+
],
|
106 |
+
"hf_avail_splits": [
|
107 |
+
"train",
|
108 |
+
"test",
|
109 |
+
"validation"
|
110 |
+
],
|
111 |
+
"evaluation_splits": [
|
112 |
+
"test"
|
113 |
+
],
|
114 |
+
"few_shots_split": null,
|
115 |
+
"few_shots_select": null,
|
116 |
+
"generation_size": 2048,
|
117 |
+
"stop_sequence": [
|
118 |
+
"\n"
|
119 |
+
],
|
120 |
+
"output_regex": null,
|
121 |
+
"frozen": false,
|
122 |
+
"suite": [
|
123 |
+
"lighteval",
|
124 |
+
"math"
|
125 |
+
],
|
126 |
+
"original_num_docs": 474,
|
127 |
+
"effective_num_docs": 10,
|
128 |
+
"trust_dataset": true,
|
129 |
+
"must_remove_duplicate_docs": null
|
130 |
+
},
|
131 |
+
"lighteval|math:geometry": {
|
132 |
+
"name": "math:geometry",
|
133 |
+
"prompt_function": "math",
|
134 |
+
"hf_repo": "lighteval/MATH",
|
135 |
+
"hf_subset": "geometry",
|
136 |
+
"metric": [
|
137 |
+
"quasi_exact_match_math"
|
138 |
+
],
|
139 |
+
"hf_avail_splits": [
|
140 |
+
"train",
|
141 |
+
"test",
|
142 |
+
"validation"
|
143 |
+
],
|
144 |
+
"evaluation_splits": [
|
145 |
+
"test"
|
146 |
+
],
|
147 |
+
"few_shots_split": null,
|
148 |
+
"few_shots_select": null,
|
149 |
+
"generation_size": 2048,
|
150 |
+
"stop_sequence": [
|
151 |
+
"\n"
|
152 |
+
],
|
153 |
+
"output_regex": null,
|
154 |
+
"frozen": false,
|
155 |
+
"suite": [
|
156 |
+
"lighteval",
|
157 |
+
"math"
|
158 |
+
],
|
159 |
+
"original_num_docs": 479,
|
160 |
+
"effective_num_docs": 10,
|
161 |
+
"trust_dataset": true,
|
162 |
+
"must_remove_duplicate_docs": null
|
163 |
+
},
|
164 |
+
"lighteval|math:intermediate_algebra": {
|
165 |
+
"name": "math:intermediate_algebra",
|
166 |
+
"prompt_function": "math",
|
167 |
+
"hf_repo": "lighteval/MATH",
|
168 |
+
"hf_subset": "intermediate_algebra",
|
169 |
+
"metric": [
|
170 |
+
"quasi_exact_match_math"
|
171 |
+
],
|
172 |
+
"hf_avail_splits": [
|
173 |
+
"train",
|
174 |
+
"test",
|
175 |
+
"validation"
|
176 |
+
],
|
177 |
+
"evaluation_splits": [
|
178 |
+
"test"
|
179 |
+
],
|
180 |
+
"few_shots_split": null,
|
181 |
+
"few_shots_select": null,
|
182 |
+
"generation_size": 2048,
|
183 |
+
"stop_sequence": [
|
184 |
+
"\n"
|
185 |
+
],
|
186 |
+
"output_regex": null,
|
187 |
+
"frozen": false,
|
188 |
+
"suite": [
|
189 |
+
"lighteval",
|
190 |
+
"math"
|
191 |
+
],
|
192 |
+
"original_num_docs": 903,
|
193 |
+
"effective_num_docs": 10,
|
194 |
+
"trust_dataset": true,
|
195 |
+
"must_remove_duplicate_docs": null
|
196 |
+
},
|
197 |
+
"lighteval|math:number_theory": {
|
198 |
+
"name": "math:number_theory",
|
199 |
+
"prompt_function": "math",
|
200 |
+
"hf_repo": "lighteval/MATH",
|
201 |
+
"hf_subset": "number_theory",
|
202 |
+
"metric": [
|
203 |
+
"quasi_exact_match_math"
|
204 |
+
],
|
205 |
+
"hf_avail_splits": [
|
206 |
+
"train",
|
207 |
+
"test",
|
208 |
+
"validation"
|
209 |
+
],
|
210 |
+
"evaluation_splits": [
|
211 |
+
"test"
|
212 |
+
],
|
213 |
+
"few_shots_split": null,
|
214 |
+
"few_shots_select": null,
|
215 |
+
"generation_size": 2048,
|
216 |
+
"stop_sequence": [
|
217 |
+
"\n"
|
218 |
+
],
|
219 |
+
"output_regex": null,
|
220 |
+
"frozen": false,
|
221 |
+
"suite": [
|
222 |
+
"lighteval",
|
223 |
+
"math"
|
224 |
+
],
|
225 |
+
"original_num_docs": 540,
|
226 |
+
"effective_num_docs": 10,
|
227 |
+
"trust_dataset": true,
|
228 |
+
"must_remove_duplicate_docs": null
|
229 |
+
},
|
230 |
+
"lighteval|math:prealgebra": {
|
231 |
+
"name": "math:prealgebra",
|
232 |
+
"prompt_function": "math",
|
233 |
+
"hf_repo": "lighteval/MATH",
|
234 |
+
"hf_subset": "prealgebra",
|
235 |
+
"metric": [
|
236 |
+
"quasi_exact_match_math"
|
237 |
+
],
|
238 |
+
"hf_avail_splits": [
|
239 |
+
"train",
|
240 |
+
"test",
|
241 |
+
"validation"
|
242 |
+
],
|
243 |
+
"evaluation_splits": [
|
244 |
+
"test"
|
245 |
+
],
|
246 |
+
"few_shots_split": null,
|
247 |
+
"few_shots_select": null,
|
248 |
+
"generation_size": 2048,
|
249 |
+
"stop_sequence": [
|
250 |
+
"\n"
|
251 |
+
],
|
252 |
+
"output_regex": null,
|
253 |
+
"frozen": false,
|
254 |
+
"suite": [
|
255 |
+
"lighteval",
|
256 |
+
"math"
|
257 |
+
],
|
258 |
+
"original_num_docs": 871,
|
259 |
+
"effective_num_docs": 10,
|
260 |
+
"trust_dataset": true,
|
261 |
+
"must_remove_duplicate_docs": null
|
262 |
+
},
|
263 |
+
"lighteval|math:precalculus": {
|
264 |
+
"name": "math:precalculus",
|
265 |
+
"prompt_function": "math",
|
266 |
+
"hf_repo": "lighteval/MATH",
|
267 |
+
"hf_subset": "precalculus",
|
268 |
+
"metric": [
|
269 |
+
"quasi_exact_match_math"
|
270 |
+
],
|
271 |
+
"hf_avail_splits": [
|
272 |
+
"train",
|
273 |
+
"test",
|
274 |
+
"validation"
|
275 |
+
],
|
276 |
+
"evaluation_splits": [
|
277 |
+
"test"
|
278 |
+
],
|
279 |
+
"few_shots_split": null,
|
280 |
+
"few_shots_select": null,
|
281 |
+
"generation_size": 2048,
|
282 |
+
"stop_sequence": [
|
283 |
+
"\n"
|
284 |
+
],
|
285 |
+
"output_regex": null,
|
286 |
+
"frozen": false,
|
287 |
+
"suite": [
|
288 |
+
"lighteval",
|
289 |
+
"math"
|
290 |
+
],
|
291 |
+
"original_num_docs": 546,
|
292 |
+
"effective_num_docs": 10,
|
293 |
+
"trust_dataset": true,
|
294 |
+
"must_remove_duplicate_docs": null
|
295 |
+
}
|
296 |
+
},
|
297 |
+
"summary_tasks": {
|
298 |
+
"lighteval|math:algebra|0": {
|
299 |
+
"hashes": {
|
300 |
+
"hash_examples": "a13d68854ca927ce",
|
301 |
+
"hash_full_prompts": "7e0d2b25e14caad6",
|
302 |
+
"hash_input_tokens": "58c8826560827dfc",
|
303 |
+
"hash_cont_tokens": "310523472b3267fc"
|
304 |
+
},
|
305 |
+
"truncated": 10,
|
306 |
+
"non_truncated": 0,
|
307 |
+
"padded": 3,
|
308 |
+
"non_padded": 7,
|
309 |
+
"effective_few_shots": 0.0,
|
310 |
+
"num_truncated_few_shots": 0
|
311 |
+
},
|
312 |
+
"lighteval|math:counting_and_probability|0": {
|
313 |
+
"hashes": {
|
314 |
+
"hash_examples": "a8004c36a2d9cb68",
|
315 |
+
"hash_full_prompts": "2acaf205499ed79c",
|
316 |
+
"hash_input_tokens": "d76018dbed1fcc48",
|
317 |
+
"hash_cont_tokens": "733b9c4cba844ec2"
|
318 |
+
},
|
319 |
+
"truncated": 10,
|
320 |
+
"non_truncated": 0,
|
321 |
+
"padded": 1,
|
322 |
+
"non_padded": 9,
|
323 |
+
"effective_few_shots": 0.0,
|
324 |
+
"num_truncated_few_shots": 0
|
325 |
+
},
|
326 |
+
"lighteval|math:geometry|0": {
|
327 |
+
"hashes": {
|
328 |
+
"hash_examples": "5e12e37f7378cc4c",
|
329 |
+
"hash_full_prompts": "32e7c26bfc66828d",
|
330 |
+
"hash_input_tokens": "288b95f1a6a9a3d5",
|
331 |
+
"hash_cont_tokens": "8d9422af27507fe7"
|
332 |
+
},
|
333 |
+
"truncated": 10,
|
334 |
+
"non_truncated": 0,
|
335 |
+
"padded": 5,
|
336 |
+
"non_padded": 5,
|
337 |
+
"effective_few_shots": 0.0,
|
338 |
+
"num_truncated_few_shots": 0
|
339 |
+
},
|
340 |
+
"lighteval|math:intermediate_algebra|0": {
|
341 |
+
"hashes": {
|
342 |
+
"hash_examples": "71738fc49d471d6d",
|
343 |
+
"hash_full_prompts": "6779f5c079af81a6",
|
344 |
+
"hash_input_tokens": "53a5702086e49106",
|
345 |
+
"hash_cont_tokens": "6f8722f0a58ef37a"
|
346 |
+
},
|
347 |
+
"truncated": 10,
|
348 |
+
"non_truncated": 0,
|
349 |
+
"padded": 3,
|
350 |
+
"non_padded": 7,
|
351 |
+
"effective_few_shots": 0.0,
|
352 |
+
"num_truncated_few_shots": 0
|
353 |
+
},
|
354 |
+
"lighteval|math:number_theory|0": {
|
355 |
+
"hashes": {
|
356 |
+
"hash_examples": "bdb66471a0eed93a",
|
357 |
+
"hash_full_prompts": "9985b650f03b8f91",
|
358 |
+
"hash_input_tokens": "8b240f273f300e85",
|
359 |
+
"hash_cont_tokens": "ae8bd3b0e9f74ac5"
|
360 |
+
},
|
361 |
+
"truncated": 10,
|
362 |
+
"non_truncated": 0,
|
363 |
+
"padded": 5,
|
364 |
+
"non_padded": 5,
|
365 |
+
"effective_few_shots": 0.0,
|
366 |
+
"num_truncated_few_shots": 0
|
367 |
+
},
|
368 |
+
"lighteval|math:prealgebra|0": {
|
369 |
+
"hashes": {
|
370 |
+
"hash_examples": "3c59373ec7e3a94a",
|
371 |
+
"hash_full_prompts": "722cea4098cecd00",
|
372 |
+
"hash_input_tokens": "6788548239da9c91",
|
373 |
+
"hash_cont_tokens": "c2f2eaca08a1e171"
|
374 |
+
},
|
375 |
+
"truncated": 10,
|
376 |
+
"non_truncated": 0,
|
377 |
+
"padded": 1,
|
378 |
+
"non_padded": 9,
|
379 |
+
"effective_few_shots": 0.0,
|
380 |
+
"num_truncated_few_shots": 0
|
381 |
+
},
|
382 |
+
"lighteval|math:precalculus|0": {
|
383 |
+
"hashes": {
|
384 |
+
"hash_examples": "8a97d7d7bd780ca3",
|
385 |
+
"hash_full_prompts": "df2793e826f0dfcc",
|
386 |
+
"hash_input_tokens": "90b2332eeca3284b",
|
387 |
+
"hash_cont_tokens": "56aff6344490817d"
|
388 |
+
},
|
389 |
+
"truncated": 10,
|
390 |
+
"non_truncated": 0,
|
391 |
+
"padded": 4,
|
392 |
+
"non_padded": 6,
|
393 |
+
"effective_few_shots": 0.0,
|
394 |
+
"num_truncated_few_shots": 0
|
395 |
+
}
|
396 |
+
},
|
397 |
+
"summary_general": {
|
398 |
+
"hashes": {
|
399 |
+
"hash_examples": "e05a305d27ed0540",
|
400 |
+
"hash_full_prompts": "7847980860d96b71",
|
401 |
+
"hash_input_tokens": "a2515743ef01f5cf",
|
402 |
+
"hash_cont_tokens": "734c8abceb12e43b"
|
403 |
+
},
|
404 |
+
"truncated": 70,
|
405 |
+
"non_truncated": 0,
|
406 |
+
"padded": 22,
|
407 |
+
"non_padded": 48,
|
408 |
+
"num_truncated_few_shots": 0
|
409 |
+
}
|
410 |
+
}
|