Upload eval_results/AI-MO/Eurus-7b-sft/aimo_v01.00/mini_math_v2_cot/results_2024-04-29T13-29-15.050137.json with huggingface_hub
Browse files
eval_results/AI-MO/Eurus-7b-sft/aimo_v01.00/mini_math_v2_cot/results_2024-04-29T13-29-15.050137.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": 391563.467611792,
|
9 |
+
"end_time": 392527.973125356,
|
10 |
+
"total_evaluation_time_secondes": "964.5055135640432",
|
11 |
+
"model_name": "AI-MO/Eurus-7b-sft",
|
12 |
+
"model_sha": "c85aa9511e4c7a9f5f061d5d8e2016536482f006",
|
13 |
+
"model_dtype": "torch.bfloat16",
|
14 |
+
"model_size": "13.99 GB",
|
15 |
+
"config": null
|
16 |
+
},
|
17 |
+
"results": {
|
18 |
+
"custom|mini_math_v2_cot:level_1|0": {
|
19 |
+
"qem": 0.34285714285714286,
|
20 |
+
"qem_stderr": 0.08140424227436863
|
21 |
+
},
|
22 |
+
"custom|mini_math_v2_cot:level_2|0": {
|
23 |
+
"qem": 0.3194444444444444,
|
24 |
+
"qem_stderr": 0.05533504751887217
|
25 |
+
},
|
26 |
+
"custom|mini_math_v2_cot:level_3|0": {
|
27 |
+
"qem": 0.16666666666666666,
|
28 |
+
"qem_stderr": 0.03950378859499813
|
29 |
+
},
|
30 |
+
"custom|mini_math_v2_cot:level_4|0": {
|
31 |
+
"qem": 0.10309278350515463,
|
32 |
+
"qem_stderr": 0.031035037574206968
|
33 |
+
},
|
34 |
+
"custom|mini_math_v2_cot:level_5|0": {
|
35 |
+
"qem": 0.03773584905660377,
|
36 |
+
"qem_stderr": 0.01859642944545588
|
37 |
+
},
|
38 |
+
"custom|mini_math_v2_cot:_average|0": {
|
39 |
+
"qem": 0.19395937730600246,
|
40 |
+
"qem_stderr": 0.04517490908158036
|
41 |
+
},
|
42 |
+
"all": {
|
43 |
+
"qem": 0.19395937730600246,
|
44 |
+
"qem_stderr": 0.04517490908158036
|
45 |
+
}
|
46 |
+
},
|
47 |
+
"versions": {
|
48 |
+
"custom|mini_math_v2_cot:level_1|0": 0,
|
49 |
+
"custom|mini_math_v2_cot:level_2|0": 0,
|
50 |
+
"custom|mini_math_v2_cot:level_3|0": 0,
|
51 |
+
"custom|mini_math_v2_cot:level_4|0": 0,
|
52 |
+
"custom|mini_math_v2_cot:level_5|0": 0
|
53 |
+
},
|
54 |
+
"config_tasks": {
|
55 |
+
"custom|mini_math_v2_cot:level_1": {
|
56 |
+
"name": "mini_math_v2_cot:level_1",
|
57 |
+
"prompt_function": "minimath_prompt_fn",
|
58 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
59 |
+
"hf_subset": "Level 1",
|
60 |
+
"metric": [
|
61 |
+
"quasi_exact_match_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_cot:level_2": {
|
85 |
+
"name": "mini_math_v2_cot:level_2",
|
86 |
+
"prompt_function": "minimath_prompt_fn",
|
87 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
88 |
+
"hf_subset": "Level 2",
|
89 |
+
"metric": [
|
90 |
+
"quasi_exact_match_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_cot:level_3": {
|
114 |
+
"name": "mini_math_v2_cot:level_3",
|
115 |
+
"prompt_function": "minimath_prompt_fn",
|
116 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
117 |
+
"hf_subset": "Level 3",
|
118 |
+
"metric": [
|
119 |
+
"quasi_exact_match_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_cot:level_4": {
|
143 |
+
"name": "mini_math_v2_cot:level_4",
|
144 |
+
"prompt_function": "minimath_prompt_fn",
|
145 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
146 |
+
"hf_subset": "Level 4",
|
147 |
+
"metric": [
|
148 |
+
"quasi_exact_match_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_cot:level_5": {
|
172 |
+
"name": "mini_math_v2_cot:level_5",
|
173 |
+
"prompt_function": "minimath_prompt_fn",
|
174 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
175 |
+
"hf_subset": "Level 5",
|
176 |
+
"metric": [
|
177 |
+
"quasi_exact_match_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_cot:level_1|0": {
|
203 |
+
"hashes": {
|
204 |
+
"hash_examples": "6e46c83c3c04fd6a",
|
205 |
+
"hash_full_prompts": "f82f7f6570e0d1d7",
|
206 |
+
"hash_input_tokens": "c337b401eb02a335",
|
207 |
+
"hash_cont_tokens": "7b1537dff805f854"
|
208 |
+
},
|
209 |
+
"truncated": 35,
|
210 |
+
"non_truncated": 0,
|
211 |
+
"padded": 23,
|
212 |
+
"non_padded": 12,
|
213 |
+
"effective_few_shots": 0.0,
|
214 |
+
"num_truncated_few_shots": 0
|
215 |
+
},
|
216 |
+
"custom|mini_math_v2_cot:level_2|0": {
|
217 |
+
"hashes": {
|
218 |
+
"hash_examples": "889713509c9bbbc1",
|
219 |
+
"hash_full_prompts": "70d857ef2d839ba0",
|
220 |
+
"hash_input_tokens": "7a576f7464b4081a",
|
221 |
+
"hash_cont_tokens": "3d333a27a2a927c9"
|
222 |
+
},
|
223 |
+
"truncated": 71,
|
224 |
+
"non_truncated": 1,
|
225 |
+
"padded": 38,
|
226 |
+
"non_padded": 34,
|
227 |
+
"effective_few_shots": 0.0,
|
228 |
+
"num_truncated_few_shots": 0
|
229 |
+
},
|
230 |
+
"custom|mini_math_v2_cot:level_3|0": {
|
231 |
+
"hashes": {
|
232 |
+
"hash_examples": "325e5cc88f99c065",
|
233 |
+
"hash_full_prompts": "ac84f6c9c8b958ae",
|
234 |
+
"hash_input_tokens": "68ee20cb97b0e202",
|
235 |
+
"hash_cont_tokens": "8be45f212900c86c"
|
236 |
+
},
|
237 |
+
"truncated": 90,
|
238 |
+
"non_truncated": 0,
|
239 |
+
"padded": 33,
|
240 |
+
"non_padded": 57,
|
241 |
+
"effective_few_shots": 0.0,
|
242 |
+
"num_truncated_few_shots": 0
|
243 |
+
},
|
244 |
+
"custom|mini_math_v2_cot:level_4|0": {
|
245 |
+
"hashes": {
|
246 |
+
"hash_examples": "797712bab625f507",
|
247 |
+
"hash_full_prompts": "b76096d7d92e2296",
|
248 |
+
"hash_input_tokens": "7de858f872668166",
|
249 |
+
"hash_cont_tokens": "3fe29b7d95997781"
|
250 |
+
},
|
251 |
+
"truncated": 97,
|
252 |
+
"non_truncated": 0,
|
253 |
+
"padded": 38,
|
254 |
+
"non_padded": 59,
|
255 |
+
"effective_few_shots": 0.0,
|
256 |
+
"num_truncated_few_shots": 0
|
257 |
+
},
|
258 |
+
"custom|mini_math_v2_cot:level_5|0": {
|
259 |
+
"hashes": {
|
260 |
+
"hash_examples": "d263622047b94222",
|
261 |
+
"hash_full_prompts": "b38ec0550a8d3581",
|
262 |
+
"hash_input_tokens": "4eea4ab5411e66e7",
|
263 |
+
"hash_cont_tokens": "2cb3146d0080f788"
|
264 |
+
},
|
265 |
+
"truncated": 106,
|
266 |
+
"non_truncated": 0,
|
267 |
+
"padded": 32,
|
268 |
+
"non_padded": 74,
|
269 |
+
"effective_few_shots": 0.0,
|
270 |
+
"num_truncated_few_shots": 0
|
271 |
+
}
|
272 |
+
},
|
273 |
+
"summary_general": {
|
274 |
+
"hashes": {
|
275 |
+
"hash_examples": "3cf20f18ddb05dd8",
|
276 |
+
"hash_full_prompts": "b74e3580ef05e271",
|
277 |
+
"hash_input_tokens": "3f1e669b8185c08b",
|
278 |
+
"hash_cont_tokens": "ab1a8b9154fbc47c"
|
279 |
+
},
|
280 |
+
"truncated": 399,
|
281 |
+
"non_truncated": 1,
|
282 |
+
"padded": 164,
|
283 |
+
"non_padded": 236,
|
284 |
+
"num_truncated_few_shots": 0
|
285 |
+
}
|
286 |
+
}
|