Upload eval_results/Qwen/Qwen2.5-Math-1.5B-Instruct/main/mini_math/results_2025-01-23T09-04-03.274525.json with huggingface_hub
Browse files
eval_results/Qwen/Qwen2.5-Math-1.5B-Instruct/main/mini_math/results_2025-01-23T09-04-03.274525.json
ADDED
@@ -0,0 +1,340 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"config_general": {
|
3 |
+
"lighteval_sha": "?",
|
4 |
+
"num_fewshot_seeds": 1,
|
5 |
+
"override_batch_size": -1,
|
6 |
+
"max_samples": null,
|
7 |
+
"job_id": 0,
|
8 |
+
"start_time": 466297.931738236,
|
9 |
+
"end_time": 466364.583520329,
|
10 |
+
"total_evaluation_time_secondes": "66.65178209298756",
|
11 |
+
"model_name": "Qwen/Qwen2.5-Math-1.5B-Instruct",
|
12 |
+
"model_sha": "",
|
13 |
+
"model_dtype": null,
|
14 |
+
"model_size": null
|
15 |
+
},
|
16 |
+
"results": {
|
17 |
+
"custom|mini_math:level_1|0": {
|
18 |
+
"qem": 0.8571428571428571,
|
19 |
+
"qem_stderr": 0.060012003601200396
|
20 |
+
},
|
21 |
+
"custom|mini_math:level_2|0": {
|
22 |
+
"qem": 0.8611111111111112,
|
23 |
+
"qem_stderr": 0.04104253469235572
|
24 |
+
},
|
25 |
+
"custom|mini_math:level_3|0": {
|
26 |
+
"qem": 0.8777777777777778,
|
27 |
+
"qem_stderr": 0.03471941311982698
|
28 |
+
},
|
29 |
+
"custom|mini_math:level_4|0": {
|
30 |
+
"qem": 0.711340206185567,
|
31 |
+
"qem_stderr": 0.04624834650754217
|
32 |
+
},
|
33 |
+
"custom|mini_math:level_5|0": {
|
34 |
+
"qem": 0.42452830188679247,
|
35 |
+
"qem_stderr": 0.04823593037243471
|
36 |
+
},
|
37 |
+
"custom|mini_math:_average|0": {
|
38 |
+
"qem": 0.7463800508208212,
|
39 |
+
"qem_stderr": 0.046051645658671994
|
40 |
+
},
|
41 |
+
"all": {
|
42 |
+
"qem": 0.7463800508208212,
|
43 |
+
"qem_stderr": 0.046051645658671994
|
44 |
+
}
|
45 |
+
},
|
46 |
+
"versions": {
|
47 |
+
"custom|mini_math:level_1|0": 0,
|
48 |
+
"custom|mini_math:level_2|0": 0,
|
49 |
+
"custom|mini_math:level_3|0": 0,
|
50 |
+
"custom|mini_math:level_4|0": 0,
|
51 |
+
"custom|mini_math:level_5|0": 0
|
52 |
+
},
|
53 |
+
"config_tasks": {
|
54 |
+
"custom|mini_math:level_1": {
|
55 |
+
"name": "mini_math:level_1",
|
56 |
+
"prompt_function": "math",
|
57 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
58 |
+
"hf_subset": "Level 1",
|
59 |
+
"metric": [
|
60 |
+
{
|
61 |
+
"metric_name": "qem",
|
62 |
+
"higher_is_better": true,
|
63 |
+
"category": "3",
|
64 |
+
"use_case": "5",
|
65 |
+
"sample_level_fn": "compute",
|
66 |
+
"corpus_level_fn": "mean"
|
67 |
+
}
|
68 |
+
],
|
69 |
+
"hf_revision": null,
|
70 |
+
"hf_filter": null,
|
71 |
+
"hf_avail_splits": [
|
72 |
+
"train",
|
73 |
+
"test"
|
74 |
+
],
|
75 |
+
"trust_dataset": false,
|
76 |
+
"evaluation_splits": [
|
77 |
+
"test"
|
78 |
+
],
|
79 |
+
"few_shots_split": null,
|
80 |
+
"few_shots_select": null,
|
81 |
+
"generation_size": 2048,
|
82 |
+
"generation_grammar": null,
|
83 |
+
"stop_sequence": [],
|
84 |
+
"num_samples": null,
|
85 |
+
"suite": [
|
86 |
+
"custom",
|
87 |
+
"mini_math"
|
88 |
+
],
|
89 |
+
"original_num_docs": 35,
|
90 |
+
"effective_num_docs": 35,
|
91 |
+
"must_remove_duplicate_docs": false,
|
92 |
+
"version": 0
|
93 |
+
},
|
94 |
+
"custom|mini_math:level_2": {
|
95 |
+
"name": "mini_math:level_2",
|
96 |
+
"prompt_function": "math",
|
97 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
98 |
+
"hf_subset": "Level 2",
|
99 |
+
"metric": [
|
100 |
+
{
|
101 |
+
"metric_name": "qem",
|
102 |
+
"higher_is_better": true,
|
103 |
+
"category": "3",
|
104 |
+
"use_case": "5",
|
105 |
+
"sample_level_fn": "compute",
|
106 |
+
"corpus_level_fn": "mean"
|
107 |
+
}
|
108 |
+
],
|
109 |
+
"hf_revision": null,
|
110 |
+
"hf_filter": null,
|
111 |
+
"hf_avail_splits": [
|
112 |
+
"train",
|
113 |
+
"test"
|
114 |
+
],
|
115 |
+
"trust_dataset": false,
|
116 |
+
"evaluation_splits": [
|
117 |
+
"test"
|
118 |
+
],
|
119 |
+
"few_shots_split": null,
|
120 |
+
"few_shots_select": null,
|
121 |
+
"generation_size": 2048,
|
122 |
+
"generation_grammar": null,
|
123 |
+
"stop_sequence": [],
|
124 |
+
"num_samples": null,
|
125 |
+
"suite": [
|
126 |
+
"custom",
|
127 |
+
"mini_math"
|
128 |
+
],
|
129 |
+
"original_num_docs": 72,
|
130 |
+
"effective_num_docs": 72,
|
131 |
+
"must_remove_duplicate_docs": false,
|
132 |
+
"version": 0
|
133 |
+
},
|
134 |
+
"custom|mini_math:level_3": {
|
135 |
+
"name": "mini_math:level_3",
|
136 |
+
"prompt_function": "math",
|
137 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
138 |
+
"hf_subset": "Level 3",
|
139 |
+
"metric": [
|
140 |
+
{
|
141 |
+
"metric_name": "qem",
|
142 |
+
"higher_is_better": true,
|
143 |
+
"category": "3",
|
144 |
+
"use_case": "5",
|
145 |
+
"sample_level_fn": "compute",
|
146 |
+
"corpus_level_fn": "mean"
|
147 |
+
}
|
148 |
+
],
|
149 |
+
"hf_revision": null,
|
150 |
+
"hf_filter": null,
|
151 |
+
"hf_avail_splits": [
|
152 |
+
"train",
|
153 |
+
"test"
|
154 |
+
],
|
155 |
+
"trust_dataset": false,
|
156 |
+
"evaluation_splits": [
|
157 |
+
"test"
|
158 |
+
],
|
159 |
+
"few_shots_split": null,
|
160 |
+
"few_shots_select": null,
|
161 |
+
"generation_size": 2048,
|
162 |
+
"generation_grammar": null,
|
163 |
+
"stop_sequence": [],
|
164 |
+
"num_samples": null,
|
165 |
+
"suite": [
|
166 |
+
"custom",
|
167 |
+
"mini_math"
|
168 |
+
],
|
169 |
+
"original_num_docs": 90,
|
170 |
+
"effective_num_docs": 90,
|
171 |
+
"must_remove_duplicate_docs": false,
|
172 |
+
"version": 0
|
173 |
+
},
|
174 |
+
"custom|mini_math:level_4": {
|
175 |
+
"name": "mini_math:level_4",
|
176 |
+
"prompt_function": "math",
|
177 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
178 |
+
"hf_subset": "Level 4",
|
179 |
+
"metric": [
|
180 |
+
{
|
181 |
+
"metric_name": "qem",
|
182 |
+
"higher_is_better": true,
|
183 |
+
"category": "3",
|
184 |
+
"use_case": "5",
|
185 |
+
"sample_level_fn": "compute",
|
186 |
+
"corpus_level_fn": "mean"
|
187 |
+
}
|
188 |
+
],
|
189 |
+
"hf_revision": null,
|
190 |
+
"hf_filter": null,
|
191 |
+
"hf_avail_splits": [
|
192 |
+
"train",
|
193 |
+
"test"
|
194 |
+
],
|
195 |
+
"trust_dataset": false,
|
196 |
+
"evaluation_splits": [
|
197 |
+
"test"
|
198 |
+
],
|
199 |
+
"few_shots_split": null,
|
200 |
+
"few_shots_select": null,
|
201 |
+
"generation_size": 2048,
|
202 |
+
"generation_grammar": null,
|
203 |
+
"stop_sequence": [],
|
204 |
+
"num_samples": null,
|
205 |
+
"suite": [
|
206 |
+
"custom",
|
207 |
+
"mini_math"
|
208 |
+
],
|
209 |
+
"original_num_docs": 97,
|
210 |
+
"effective_num_docs": 97,
|
211 |
+
"must_remove_duplicate_docs": false,
|
212 |
+
"version": 0
|
213 |
+
},
|
214 |
+
"custom|mini_math:level_5": {
|
215 |
+
"name": "mini_math:level_5",
|
216 |
+
"prompt_function": "math",
|
217 |
+
"hf_repo": "AI-MO/lighteval-mini-math",
|
218 |
+
"hf_subset": "Level 5",
|
219 |
+
"metric": [
|
220 |
+
{
|
221 |
+
"metric_name": "qem",
|
222 |
+
"higher_is_better": true,
|
223 |
+
"category": "3",
|
224 |
+
"use_case": "5",
|
225 |
+
"sample_level_fn": "compute",
|
226 |
+
"corpus_level_fn": "mean"
|
227 |
+
}
|
228 |
+
],
|
229 |
+
"hf_revision": null,
|
230 |
+
"hf_filter": null,
|
231 |
+
"hf_avail_splits": [
|
232 |
+
"train",
|
233 |
+
"test"
|
234 |
+
],
|
235 |
+
"trust_dataset": false,
|
236 |
+
"evaluation_splits": [
|
237 |
+
"test"
|
238 |
+
],
|
239 |
+
"few_shots_split": null,
|
240 |
+
"few_shots_select": null,
|
241 |
+
"generation_size": 2048,
|
242 |
+
"generation_grammar": null,
|
243 |
+
"stop_sequence": [],
|
244 |
+
"num_samples": null,
|
245 |
+
"suite": [
|
246 |
+
"custom",
|
247 |
+
"mini_math"
|
248 |
+
],
|
249 |
+
"original_num_docs": 106,
|
250 |
+
"effective_num_docs": 106,
|
251 |
+
"must_remove_duplicate_docs": false,
|
252 |
+
"version": 0
|
253 |
+
}
|
254 |
+
},
|
255 |
+
"summary_tasks": {
|
256 |
+
"custom|mini_math:level_1|0": {
|
257 |
+
"hashes": {
|
258 |
+
"hash_examples": "f57f08853806c023",
|
259 |
+
"hash_full_prompts": "9a3a20fd93db6f19",
|
260 |
+
"hash_input_tokens": "b870f759aaf9fa2e",
|
261 |
+
"hash_cont_tokens": "13340a9b824c56a8"
|
262 |
+
},
|
263 |
+
"truncated": 0,
|
264 |
+
"non_truncated": 35,
|
265 |
+
"padded": 0,
|
266 |
+
"non_padded": 35,
|
267 |
+
"effective_few_shots": 0.0,
|
268 |
+
"num_truncated_few_shots": 0
|
269 |
+
},
|
270 |
+
"custom|mini_math:level_2|0": {
|
271 |
+
"hashes": {
|
272 |
+
"hash_examples": "b4fc7cab14532241",
|
273 |
+
"hash_full_prompts": "e9cf9a7613462711",
|
274 |
+
"hash_input_tokens": "97a7ed80eadc63e3",
|
275 |
+
"hash_cont_tokens": "cdcb1b694822ca5d"
|
276 |
+
},
|
277 |
+
"truncated": 0,
|
278 |
+
"non_truncated": 72,
|
279 |
+
"padded": 0,
|
280 |
+
"non_padded": 72,
|
281 |
+
"effective_few_shots": 0.0,
|
282 |
+
"num_truncated_few_shots": 0
|
283 |
+
},
|
284 |
+
"custom|mini_math:level_3|0": {
|
285 |
+
"hashes": {
|
286 |
+
"hash_examples": "07877d35e91b6109",
|
287 |
+
"hash_full_prompts": "c1f9f7bceef1a641",
|
288 |
+
"hash_input_tokens": "818a30a7ae391ca5",
|
289 |
+
"hash_cont_tokens": "aad18f03e74d477c"
|
290 |
+
},
|
291 |
+
"truncated": 0,
|
292 |
+
"non_truncated": 90,
|
293 |
+
"padded": 0,
|
294 |
+
"non_padded": 90,
|
295 |
+
"effective_few_shots": 0.0,
|
296 |
+
"num_truncated_few_shots": 0
|
297 |
+
},
|
298 |
+
"custom|mini_math:level_4|0": {
|
299 |
+
"hashes": {
|
300 |
+
"hash_examples": "016981666505c34d",
|
301 |
+
"hash_full_prompts": "e583c079179c8459",
|
302 |
+
"hash_input_tokens": "4828de070bbe53be",
|
303 |
+
"hash_cont_tokens": "29e2954e09d16e1d"
|
304 |
+
},
|
305 |
+
"truncated": 0,
|
306 |
+
"non_truncated": 97,
|
307 |
+
"padded": 0,
|
308 |
+
"non_padded": 97,
|
309 |
+
"effective_few_shots": 0.0,
|
310 |
+
"num_truncated_few_shots": 0
|
311 |
+
},
|
312 |
+
"custom|mini_math:level_5|0": {
|
313 |
+
"hashes": {
|
314 |
+
"hash_examples": "80868d5b1c260707",
|
315 |
+
"hash_full_prompts": "d45915cec626c593",
|
316 |
+
"hash_input_tokens": "5fd7765f4e8a2a9c",
|
317 |
+
"hash_cont_tokens": "7197eeac005410bb"
|
318 |
+
},
|
319 |
+
"truncated": 0,
|
320 |
+
"non_truncated": 106,
|
321 |
+
"padded": 0,
|
322 |
+
"non_padded": 106,
|
323 |
+
"effective_few_shots": 0.0,
|
324 |
+
"num_truncated_few_shots": 0
|
325 |
+
}
|
326 |
+
},
|
327 |
+
"summary_general": {
|
328 |
+
"hashes": {
|
329 |
+
"hash_examples": "e7ee3a47ca2b0866",
|
330 |
+
"hash_full_prompts": "1f7e06ec8d733918",
|
331 |
+
"hash_input_tokens": "4bcc69b39c8416ec",
|
332 |
+
"hash_cont_tokens": "dd8153c97fb558fc"
|
333 |
+
},
|
334 |
+
"truncated": 0,
|
335 |
+
"non_truncated": 400,
|
336 |
+
"padded": 0,
|
337 |
+
"non_padded": 400,
|
338 |
+
"num_truncated_few_shots": 0
|
339 |
+
}
|
340 |
+
}
|