====== Perplexity statistics ======
Mean PPL(Q)                   :   7.668204 ±   0.052714
Mean PPL(base)                :   7.669212 ±   0.052592
Cor(ln(PPL(Q)), ln(PPL(base))):  99.94%
Mean ln(PPL(Q)/PPL(base))     :  -0.000131 ±   0.000238
Mean PPL(Q)/PPL(base)         :   0.999869 ±   0.000238
Mean PPL(Q)-PPL(base)         :  -0.001008 ±   0.001825

====== KL divergence statistics ======
Mean    KLD:   0.001478 ±   0.000032
Maximum KLD:   4.769487
99.9%   KLD:   0.028515
99.0%   KLD:   0.009430
99.0%   KLD:   0.009430
Median  KLD:   0.000898
10.0%   KLD:   0.000021
 5.0%   KLD:   0.000004
 1.0%   KLD:  -0.000001
Minimum KLD:  -0.000183

====== Token probability statistics ======
Mean    Δp:  0.006 ± 0.003 %
Maximum Δp: 66.388%
99.9%   Δp:  7.271%
99.0%   Δp:  3.796%
95.0%   Δp:  1.931%
90.0%   Δp:  1.136%
75.0%   Δp:  0.236%
Median  Δp:  0.000%
25.0%   Δp: -0.229%
10.0%   Δp: -1.136%
 5.0%   Δp: -1.916%
 1.0%   Δp: -3.721%
 0.1%   Δp: -6.644%
Minimum Δp: -41.831%
RMS Δp    :  1.243 ± 0.014 %
Same top p: 97.929 ± 0.037 %