Add v3 evaluation artifact v3_reasoning_sft_manifest.json
Browse files
candidates/budgie-alignment-v2/reasoning-sft-then-rl-v3/recovery/v3_reasoning_sft_manifest.json
ADDED
|
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"scanned": 5251,
|
| 3 |
+
"verified_pool": 1600,
|
| 4 |
+
"train": 800,
|
| 5 |
+
"dev": 208,
|
| 6 |
+
"difficulty_train": {
|
| 7 |
+
"medium": 511,
|
| 8 |
+
"hard": 289
|
| 9 |
+
},
|
| 10 |
+
"sources_train": {
|
| 11 |
+
"olympiads": 357,
|
| 12 |
+
"aops_forum": 225,
|
| 13 |
+
"cn_contest": 154,
|
| 14 |
+
"amc_aime": 32,
|
| 15 |
+
"inequalities": 19,
|
| 16 |
+
"olympiads_ref": 10,
|
| 17 |
+
"number_theory": 3
|
| 18 |
+
},
|
| 19 |
+
"problem_types_train": {
|
| 20 |
+
"Number Theory": 179,
|
| 21 |
+
"Algebra": 228,
|
| 22 |
+
"Geometry": 193,
|
| 23 |
+
"Combinatorics": 119,
|
| 24 |
+
"Calculus": 19,
|
| 25 |
+
"Logic and Puzzles": 29,
|
| 26 |
+
"Other": 11,
|
| 27 |
+
"Inequalities": 22
|
| 28 |
+
},
|
| 29 |
+
"rejections": {
|
| 30 |
+
"verify": 2910,
|
| 31 |
+
"length": 740
|
| 32 |
+
},
|
| 33 |
+
"benchmark_family_rows": 0,
|
| 34 |
+
"decontamination": "normalized exact + shared fixed-gate 12-token windows + source bans",
|
| 35 |
+
"verification": "math_verify(reference solution, answer)",
|
| 36 |
+
"seconds": 69.16955494880676,
|
| 37 |
+
"difficulty_dev": {
|
| 38 |
+
"hard": 78,
|
| 39 |
+
"medium": 130
|
| 40 |
+
}
|
| 41 |
+
}
|