Upload eval_results/AI-MO/deepseek-math-7b-sft/aimo_v31.24/math_v2/results_2024-07-01T19-35-19.773558.json with huggingface_hub
Browse files
eval_results/AI-MO/deepseek-math-7b-sft/aimo_v31.24/math_v2/results_2024-07-01T19-35-19.773558.json
ADDED
@@ -0,0 +1,403 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
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": 1475617.30202661,
|
9 |
+
"end_time": 1494064.021881981,
|
10 |
+
"total_evaluation_time_secondes": "18446.719855371164",
|
11 |
+
"model_name": "AI-MO/deepseek-math-7b-sft",
|
12 |
+
"model_sha": "019bac29164fae289ce6dc9efd9edf0d576d2040",
|
13 |
+
"model_dtype": "torch.bfloat16",
|
14 |
+
"model_size": "12.87 GB",
|
15 |
+
"config": null
|
16 |
+
},
|
17 |
+
"results": {
|
18 |
+
"custom|math_v2:algebra|0": {
|
19 |
+
"qem": 0.7598989048020219,
|
20 |
+
"qem_stderr": 0.01240316825932235
|
21 |
+
},
|
22 |
+
"custom|math_v2:counting_and_probability|0": {
|
23 |
+
"qem": 0.45147679324894513,
|
24 |
+
"qem_stderr": 0.022881508288974756
|
25 |
+
},
|
26 |
+
"custom|math_v2:geometry|0": {
|
27 |
+
"qem": 0.4279749478079332,
|
28 |
+
"qem_stderr": 0.022630948181635056
|
29 |
+
},
|
30 |
+
"custom|math_v2:intermediate_algebra|0": {
|
31 |
+
"qem": 0.31007751937984496,
|
32 |
+
"qem_stderr": 0.015400400668438145
|
33 |
+
},
|
34 |
+
"custom|math_v2:number_theory|0": {
|
35 |
+
"qem": 0.5018518518518519,
|
36 |
+
"qem_stderr": 0.021536376899401997
|
37 |
+
},
|
38 |
+
"custom|math_v2:prealgebra|0": {
|
39 |
+
"qem": 0.7049368541905855,
|
40 |
+
"qem_stderr": 0.015462259793596653
|
41 |
+
},
|
42 |
+
"custom|math_v2:precalculus|0": {
|
43 |
+
"qem": 0.34065934065934067,
|
44 |
+
"qem_stderr": 0.020300971833784615
|
45 |
+
},
|
46 |
+
"custom|math_v2:_average|0": {
|
47 |
+
"qem": 0.49955374456293195,
|
48 |
+
"qem_stderr": 0.018659376275021938
|
49 |
+
},
|
50 |
+
"all": {
|
51 |
+
"qem": 0.49955374456293195,
|
52 |
+
"qem_stderr": 0.018659376275021938
|
53 |
+
}
|
54 |
+
},
|
55 |
+
"versions": {
|
56 |
+
"custom|math_v2:algebra|0": 0,
|
57 |
+
"custom|math_v2:counting_and_probability|0": 0,
|
58 |
+
"custom|math_v2:geometry|0": 0,
|
59 |
+
"custom|math_v2:intermediate_algebra|0": 0,
|
60 |
+
"custom|math_v2:number_theory|0": 0,
|
61 |
+
"custom|math_v2:prealgebra|0": 0,
|
62 |
+
"custom|math_v2:precalculus|0": 0
|
63 |
+
},
|
64 |
+
"config_tasks": {
|
65 |
+
"custom|math_v2:algebra": {
|
66 |
+
"name": "math_v2:algebra",
|
67 |
+
"prompt_function": "math_prompt_fn",
|
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": null,
|
85 |
+
"output_regex": null,
|
86 |
+
"num_samples": null,
|
87 |
+
"frozen": false,
|
88 |
+
"suite": [
|
89 |
+
"custom"
|
90 |
+
],
|
91 |
+
"original_num_docs": 1187,
|
92 |
+
"effective_num_docs": 1187,
|
93 |
+
"trust_dataset": null,
|
94 |
+
"must_remove_duplicate_docs": null,
|
95 |
+
"version": 0
|
96 |
+
},
|
97 |
+
"custom|math_v2:counting_and_probability": {
|
98 |
+
"name": "math_v2:counting_and_probability",
|
99 |
+
"prompt_function": "math_prompt_fn",
|
100 |
+
"hf_repo": "lighteval/MATH",
|
101 |
+
"hf_subset": "counting_and_probability",
|
102 |
+
"metric": [
|
103 |
+
"quasi_exact_match_math"
|
104 |
+
],
|
105 |
+
"hf_avail_splits": [
|
106 |
+
"train",
|
107 |
+
"test",
|
108 |
+
"validation"
|
109 |
+
],
|
110 |
+
"evaluation_splits": [
|
111 |
+
"test"
|
112 |
+
],
|
113 |
+
"few_shots_split": null,
|
114 |
+
"few_shots_select": null,
|
115 |
+
"generation_size": 2048,
|
116 |
+
"stop_sequence": null,
|
117 |
+
"output_regex": null,
|
118 |
+
"num_samples": null,
|
119 |
+
"frozen": false,
|
120 |
+
"suite": [
|
121 |
+
"custom"
|
122 |
+
],
|
123 |
+
"original_num_docs": 474,
|
124 |
+
"effective_num_docs": 474,
|
125 |
+
"trust_dataset": null,
|
126 |
+
"must_remove_duplicate_docs": null,
|
127 |
+
"version": 0
|
128 |
+
},
|
129 |
+
"custom|math_v2:geometry": {
|
130 |
+
"name": "math_v2:geometry",
|
131 |
+
"prompt_function": "math_prompt_fn",
|
132 |
+
"hf_repo": "lighteval/MATH",
|
133 |
+
"hf_subset": "geometry",
|
134 |
+
"metric": [
|
135 |
+
"quasi_exact_match_math"
|
136 |
+
],
|
137 |
+
"hf_avail_splits": [
|
138 |
+
"train",
|
139 |
+
"test",
|
140 |
+
"validation"
|
141 |
+
],
|
142 |
+
"evaluation_splits": [
|
143 |
+
"test"
|
144 |
+
],
|
145 |
+
"few_shots_split": null,
|
146 |
+
"few_shots_select": null,
|
147 |
+
"generation_size": 2048,
|
148 |
+
"stop_sequence": null,
|
149 |
+
"output_regex": null,
|
150 |
+
"num_samples": null,
|
151 |
+
"frozen": false,
|
152 |
+
"suite": [
|
153 |
+
"custom"
|
154 |
+
],
|
155 |
+
"original_num_docs": 479,
|
156 |
+
"effective_num_docs": 479,
|
157 |
+
"trust_dataset": null,
|
158 |
+
"must_remove_duplicate_docs": null,
|
159 |
+
"version": 0
|
160 |
+
},
|
161 |
+
"custom|math_v2:intermediate_algebra": {
|
162 |
+
"name": "math_v2:intermediate_algebra",
|
163 |
+
"prompt_function": "math_prompt_fn",
|
164 |
+
"hf_repo": "lighteval/MATH",
|
165 |
+
"hf_subset": "intermediate_algebra",
|
166 |
+
"metric": [
|
167 |
+
"quasi_exact_match_math"
|
168 |
+
],
|
169 |
+
"hf_avail_splits": [
|
170 |
+
"train",
|
171 |
+
"test",
|
172 |
+
"validation"
|
173 |
+
],
|
174 |
+
"evaluation_splits": [
|
175 |
+
"test"
|
176 |
+
],
|
177 |
+
"few_shots_split": null,
|
178 |
+
"few_shots_select": null,
|
179 |
+
"generation_size": 2048,
|
180 |
+
"stop_sequence": null,
|
181 |
+
"output_regex": null,
|
182 |
+
"num_samples": null,
|
183 |
+
"frozen": false,
|
184 |
+
"suite": [
|
185 |
+
"custom"
|
186 |
+
],
|
187 |
+
"original_num_docs": 903,
|
188 |
+
"effective_num_docs": 903,
|
189 |
+
"trust_dataset": null,
|
190 |
+
"must_remove_duplicate_docs": null,
|
191 |
+
"version": 0
|
192 |
+
},
|
193 |
+
"custom|math_v2:number_theory": {
|
194 |
+
"name": "math_v2:number_theory",
|
195 |
+
"prompt_function": "math_prompt_fn",
|
196 |
+
"hf_repo": "lighteval/MATH",
|
197 |
+
"hf_subset": "number_theory",
|
198 |
+
"metric": [
|
199 |
+
"quasi_exact_match_math"
|
200 |
+
],
|
201 |
+
"hf_avail_splits": [
|
202 |
+
"train",
|
203 |
+
"test",
|
204 |
+
"validation"
|
205 |
+
],
|
206 |
+
"evaluation_splits": [
|
207 |
+
"test"
|
208 |
+
],
|
209 |
+
"few_shots_split": null,
|
210 |
+
"few_shots_select": null,
|
211 |
+
"generation_size": 2048,
|
212 |
+
"stop_sequence": null,
|
213 |
+
"output_regex": null,
|
214 |
+
"num_samples": null,
|
215 |
+
"frozen": false,
|
216 |
+
"suite": [
|
217 |
+
"custom"
|
218 |
+
],
|
219 |
+
"original_num_docs": 540,
|
220 |
+
"effective_num_docs": 540,
|
221 |
+
"trust_dataset": null,
|
222 |
+
"must_remove_duplicate_docs": null,
|
223 |
+
"version": 0
|
224 |
+
},
|
225 |
+
"custom|math_v2:prealgebra": {
|
226 |
+
"name": "math_v2:prealgebra",
|
227 |
+
"prompt_function": "math_prompt_fn",
|
228 |
+
"hf_repo": "lighteval/MATH",
|
229 |
+
"hf_subset": "prealgebra",
|
230 |
+
"metric": [
|
231 |
+
"quasi_exact_match_math"
|
232 |
+
],
|
233 |
+
"hf_avail_splits": [
|
234 |
+
"train",
|
235 |
+
"test",
|
236 |
+
"validation"
|
237 |
+
],
|
238 |
+
"evaluation_splits": [
|
239 |
+
"test"
|
240 |
+
],
|
241 |
+
"few_shots_split": null,
|
242 |
+
"few_shots_select": null,
|
243 |
+
"generation_size": 2048,
|
244 |
+
"stop_sequence": null,
|
245 |
+
"output_regex": null,
|
246 |
+
"num_samples": null,
|
247 |
+
"frozen": false,
|
248 |
+
"suite": [
|
249 |
+
"custom"
|
250 |
+
],
|
251 |
+
"original_num_docs": 871,
|
252 |
+
"effective_num_docs": 871,
|
253 |
+
"trust_dataset": null,
|
254 |
+
"must_remove_duplicate_docs": null,
|
255 |
+
"version": 0
|
256 |
+
},
|
257 |
+
"custom|math_v2:precalculus": {
|
258 |
+
"name": "math_v2:precalculus",
|
259 |
+
"prompt_function": "math_prompt_fn",
|
260 |
+
"hf_repo": "lighteval/MATH",
|
261 |
+
"hf_subset": "precalculus",
|
262 |
+
"metric": [
|
263 |
+
"quasi_exact_match_math"
|
264 |
+
],
|
265 |
+
"hf_avail_splits": [
|
266 |
+
"train",
|
267 |
+
"test",
|
268 |
+
"validation"
|
269 |
+
],
|
270 |
+
"evaluation_splits": [
|
271 |
+
"test"
|
272 |
+
],
|
273 |
+
"few_shots_split": null,
|
274 |
+
"few_shots_select": null,
|
275 |
+
"generation_size": 2048,
|
276 |
+
"stop_sequence": null,
|
277 |
+
"output_regex": null,
|
278 |
+
"num_samples": null,
|
279 |
+
"frozen": false,
|
280 |
+
"suite": [
|
281 |
+
"custom"
|
282 |
+
],
|
283 |
+
"original_num_docs": 546,
|
284 |
+
"effective_num_docs": 546,
|
285 |
+
"trust_dataset": null,
|
286 |
+
"must_remove_duplicate_docs": null,
|
287 |
+
"version": 0
|
288 |
+
}
|
289 |
+
},
|
290 |
+
"summary_tasks": {
|
291 |
+
"custom|math_v2:algebra|0": {
|
292 |
+
"hashes": {
|
293 |
+
"hash_examples": "6ec951c5aa417d2a",
|
294 |
+
"hash_full_prompts": "4e0b4b752e408da7",
|
295 |
+
"hash_input_tokens": "bf40f4fb263422f4",
|
296 |
+
"hash_cont_tokens": "707bd8fe1c7697e3"
|
297 |
+
},
|
298 |
+
"truncated": 1187,
|
299 |
+
"non_truncated": 0,
|
300 |
+
"padded": 153,
|
301 |
+
"non_padded": 1034,
|
302 |
+
"effective_few_shots": 0.0,
|
303 |
+
"num_truncated_few_shots": 0
|
304 |
+
},
|
305 |
+
"custom|math_v2:counting_and_probability|0": {
|
306 |
+
"hashes": {
|
307 |
+
"hash_examples": "cd34cb03dc09e1ad",
|
308 |
+
"hash_full_prompts": "9ebbc169a1089204",
|
309 |
+
"hash_input_tokens": "5c7292a20350a51f",
|
310 |
+
"hash_cont_tokens": "19ec8bebe3c2c173"
|
311 |
+
},
|
312 |
+
"truncated": 474,
|
313 |
+
"non_truncated": 0,
|
314 |
+
"padded": 51,
|
315 |
+
"non_padded": 423,
|
316 |
+
"effective_few_shots": 0.0,
|
317 |
+
"num_truncated_few_shots": 0
|
318 |
+
},
|
319 |
+
"custom|math_v2:geometry|0": {
|
320 |
+
"hashes": {
|
321 |
+
"hash_examples": "e1011f83d0cb54d0",
|
322 |
+
"hash_full_prompts": "e6e4d9f63cdecf28",
|
323 |
+
"hash_input_tokens": "6bddbe80e0a34826",
|
324 |
+
"hash_cont_tokens": "e122d7d179ae1437"
|
325 |
+
},
|
326 |
+
"truncated": 479,
|
327 |
+
"non_truncated": 0,
|
328 |
+
"padded": 114,
|
329 |
+
"non_padded": 365,
|
330 |
+
"effective_few_shots": 0.0,
|
331 |
+
"num_truncated_few_shots": 0
|
332 |
+
},
|
333 |
+
"custom|math_v2:intermediate_algebra|0": {
|
334 |
+
"hashes": {
|
335 |
+
"hash_examples": "aa72155be072b11c",
|
336 |
+
"hash_full_prompts": "38dad98a61e4f0ed",
|
337 |
+
"hash_input_tokens": "01f8e620b2b8b5cd",
|
338 |
+
"hash_cont_tokens": "fe36c6ced1defaae"
|
339 |
+
},
|
340 |
+
"truncated": 903,
|
341 |
+
"non_truncated": 0,
|
342 |
+
"padded": 43,
|
343 |
+
"non_padded": 860,
|
344 |
+
"effective_few_shots": 0.0,
|
345 |
+
"num_truncated_few_shots": 0
|
346 |
+
},
|
347 |
+
"custom|math_v2:number_theory|0": {
|
348 |
+
"hashes": {
|
349 |
+
"hash_examples": "b8565befcdbe9247",
|
350 |
+
"hash_full_prompts": "8ca7a4c78d6e1280",
|
351 |
+
"hash_input_tokens": "493c495f9857478b",
|
352 |
+
"hash_cont_tokens": "22c1aa7255787a1a"
|
353 |
+
},
|
354 |
+
"truncated": 537,
|
355 |
+
"non_truncated": 3,
|
356 |
+
"padded": 6,
|
357 |
+
"non_padded": 534,
|
358 |
+
"effective_few_shots": 0.0,
|
359 |
+
"num_truncated_few_shots": 0
|
360 |
+
},
|
361 |
+
"custom|math_v2:prealgebra|0": {
|
362 |
+
"hashes": {
|
363 |
+
"hash_examples": "e04d1527fe369f16",
|
364 |
+
"hash_full_prompts": "51e0c4e3d62158e4",
|
365 |
+
"hash_input_tokens": "09c1fb1af630e23b",
|
366 |
+
"hash_cont_tokens": "30160d3ad9b92248"
|
367 |
+
},
|
368 |
+
"truncated": 871,
|
369 |
+
"non_truncated": 0,
|
370 |
+
"padded": 53,
|
371 |
+
"non_padded": 818,
|
372 |
+
"effective_few_shots": 0.0,
|
373 |
+
"num_truncated_few_shots": 0
|
374 |
+
},
|
375 |
+
"custom|math_v2:precalculus|0": {
|
376 |
+
"hashes": {
|
377 |
+
"hash_examples": "97606c134f223253",
|
378 |
+
"hash_full_prompts": "2438e95fc348976e",
|
379 |
+
"hash_input_tokens": "fc41ebda0093a5c9",
|
380 |
+
"hash_cont_tokens": "00e1f0eb76b1b2b1"
|
381 |
+
},
|
382 |
+
"truncated": 546,
|
383 |
+
"non_truncated": 0,
|
384 |
+
"padded": 21,
|
385 |
+
"non_padded": 525,
|
386 |
+
"effective_few_shots": 0.0,
|
387 |
+
"num_truncated_few_shots": 0
|
388 |
+
}
|
389 |
+
},
|
390 |
+
"summary_general": {
|
391 |
+
"hashes": {
|
392 |
+
"hash_examples": "de5f0d623d1896c2",
|
393 |
+
"hash_full_prompts": "3cb45e93275657cb",
|
394 |
+
"hash_input_tokens": "019ac50da1b1eb0a",
|
395 |
+
"hash_cont_tokens": "791281e970788ee8"
|
396 |
+
},
|
397 |
+
"truncated": 4997,
|
398 |
+
"non_truncated": 3,
|
399 |
+
"padded": 441,
|
400 |
+
"non_padded": 4559,
|
401 |
+
"num_truncated_few_shots": 0
|
402 |
+
}
|
403 |
+
}
|