Upload eval/v8_results.json with huggingface_hub
Browse files- eval/v8_results.json +69 -0
eval/v8_results.json
ADDED
|
@@ -0,0 +1,69 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"always_frontier": {
|
| 3 |
+
"success": 0.8935,
|
| 4 |
+
"avg_cost": 1.0,
|
| 5 |
+
"unsafe_rate": 0.026,
|
| 6 |
+
"false_done": 0.0805,
|
| 7 |
+
"tier_dist": {
|
| 8 |
+
"4": 2000
|
| 9 |
+
}
|
| 10 |
+
},
|
| 11 |
+
"always_cheap": {
|
| 12 |
+
"success": 0.238,
|
| 13 |
+
"avg_cost": 0.04999999999999823,
|
| 14 |
+
"unsafe_rate": 0.762,
|
| 15 |
+
"false_done": 0.0,
|
| 16 |
+
"tier_dist": {
|
| 17 |
+
"1": 2000
|
| 18 |
+
}
|
| 19 |
+
},
|
| 20 |
+
"heuristic_static": {
|
| 21 |
+
"success": 0.825,
|
| 22 |
+
"avg_cost": 0.929275000000004,
|
| 23 |
+
"unsafe_rate": 0.058,
|
| 24 |
+
"false_done": 0.117,
|
| 25 |
+
"tier_dist": {
|
| 26 |
+
"4": 674,
|
| 27 |
+
"3": 646,
|
| 28 |
+
"5": 443,
|
| 29 |
+
"2": 237
|
| 30 |
+
}
|
| 31 |
+
},
|
| 32 |
+
"oracle": {
|
| 33 |
+
"success": 1.0,
|
| 34 |
+
"avg_cost": 0.46504999999999164,
|
| 35 |
+
"unsafe_rate": 0.0,
|
| 36 |
+
"false_done": 0.0,
|
| 37 |
+
"tier_dist": {
|
| 38 |
+
"3": 647,
|
| 39 |
+
"1": 476,
|
| 40 |
+
"2": 567,
|
| 41 |
+
"4": 258,
|
| 42 |
+
"5": 52
|
| 43 |
+
}
|
| 44 |
+
},
|
| 45 |
+
"v8_dynamic+ML": {
|
| 46 |
+
"success": 0.82,
|
| 47 |
+
"avg_cost": 0.922775000000004,
|
| 48 |
+
"unsafe_rate": 0.0615,
|
| 49 |
+
"false_done": 0.1185,
|
| 50 |
+
"tier_dist": {
|
| 51 |
+
"3": 754,
|
| 52 |
+
"4": 538,
|
| 53 |
+
"5": 471,
|
| 54 |
+
"2": 237
|
| 55 |
+
}
|
| 56 |
+
},
|
| 57 |
+
"v8_dynamic_only": {
|
| 58 |
+
"success": 0.82,
|
| 59 |
+
"avg_cost": 0.922775000000004,
|
| 60 |
+
"unsafe_rate": 0.0615,
|
| 61 |
+
"false_done": 0.1185,
|
| 62 |
+
"tier_dist": {
|
| 63 |
+
"3": 754,
|
| 64 |
+
"4": 538,
|
| 65 |
+
"5": 471,
|
| 66 |
+
"2": 237
|
| 67 |
+
}
|
| 68 |
+
}
|
| 69 |
+
}
|