taf-agent / data /e4_gamma /Qwen--Qwen2.5-3B_random.json
karlexmarin's picture
feat: ship paper artefacts + CLI diagnostic alongside browser tool
535348a
raw
history blame
3.05 kB
{
"model": "Qwen/Qwen2.5-3B",
"theta": 1000000,
"gamma_pred": 0.6666637444455867,
"C_theory": 9.2103,
"distances": [
10,
20,
30,
50,
100,
200,
500,
1000,
2000
],
"distances_fit": [
30,
50,
100,
200,
500,
1000,
2000
],
"d_min_fit": 30,
"n_prompts_per_distance": 150,
"seeds": [
42,
123,
7
],
"attn_by_distance": {
"10": {
"mean_across_seeds": 0.011255048217458858,
"std_across_seeds": 0.00029170327713132915,
"per_seed_means": [
0.011365672556372981,
0.010855558921272556,
0.011543913174731036
]
},
"20": {
"mean_across_seeds": 0.008040837371307943,
"std_across_seeds": 5.735484570626992e-05,
"per_seed_means": [
0.008052886674801508,
0.00796534705751886,
0.00810427838160346
]
},
"30": {
"mean_across_seeds": 0.005607237855696843,
"std_across_seeds": 8.863202668815811e-05,
"per_seed_means": [
0.005691748578877499,
0.005645150489484271,
0.0054848144987287625
]
},
"50": {
"mean_across_seeds": 0.004305336697854929,
"std_across_seeds": 0.00013922616269980767,
"per_seed_means": [
0.004180670159403235,
0.004235686721124996,
0.004499653213036557
]
},
"100": {
"mean_across_seeds": 0.0017492879890940256,
"std_across_seeds": 0.00013272472868903432,
"per_seed_means": [
0.0016116862614095832,
0.0017075307061895728,
0.0019286469996829207
]
},
"200": {
"mean_across_seeds": 0.0007549225819174252,
"std_across_seeds": 7.111260830671966e-05,
"per_seed_means": [
0.0007408848872485881,
0.0006756992589604731,
0.0008481835995432144
]
},
"500": {
"mean_across_seeds": 0.0003144526818747787,
"std_across_seeds": 5.0788319188644444e-05,
"per_seed_means": [
0.00028346924572057714,
0.00038606213388751105,
0.00027382666601624803
]
},
"1000": {
"mean_across_seeds": 0.0002920797811627078,
"std_across_seeds": 3.889070069088199e-05,
"per_seed_means": [
0.00030768415017519147,
0.0003299515241330179,
0.0002386036691799139
]
},
"2000": {
"mean_across_seeds": 8.988117967318329e-05,
"std_across_seeds": 5.94207520177485e-06,
"per_seed_means": [
9.827569079789101e-05,
8.601787588607598e-05,
8.534997233558291e-05
]
}
},
"fit_power_law": {
"gamma": 0.9643628833140916,
"log_A": -1.871616966615359,
"R2": 0.97891,
"n_points": 7
},
"fit_exponential": {
"lambda": 0.0018280361467505031,
"log_A": -6.085205860136052,
"R2": 0.758046,
"n_points": 7
},
"delta_R2_power_minus_exp": 0.2209,
"gamma_CI_95": {
"lo": 0.8066661732202158,
"hi": 1.1104644097863652
},
"decision": "REFUTED: C not constant across \u03b8"
}