Role-Filler Attention / results.json
Failed on benchmark
1{
2 "seed": 2951,
3 "math_check": {
4 "max_role_contraction_abs_error": 1.3322676295501878e-15,
5 "exact_query_top1": true,
6 "exact_query_probability": 0.9999999999656395,
7 "exact_rebinding_l2_error": 1.161975694124784e-10,
8 "exact_entropy": 8.795314964741619e-10,
9 "mismatched_role_entropy": 0.0008730161579515994
10 },
11 "benchmark": {
12 "structured_acc": 0.995,
13 "dense_acc": 0.579,
14 "structured_err": 0.03480833023786545,
15 "dense_err": 2.383237838745117,
16 "structured_entropy": 0.005550319146098912,
17 "dense_entropy": 0.06881254905095009,
18 "wrong_entropy": 0.20854220871185467
19 },
20 "benchmark_repeat": {
21 "structured_acc": 0.995,
22 "dense_acc": 0.551,
23 "structured_err": 0.043424125760793686,
24 "dense_err": 2.5534236431121826,
25 "structured_entropy": 0.009825814736692964,
26 "dense_entropy": 0.07332719177381702,
27 "wrong_entropy": 0.180572374217896
28 }
29}