====== Perplexity statistics ======
Mean PPL(Q)                   :   8.949256 ±   0.069263
Mean PPL(base)                :   8.445938 ±   0.065177
Cor(ln(PPL(Q)), ln(PPL(base))):  97.57%
Mean ln(PPL(Q)/PPL(base))     :   0.057885 ±   0.001704
Mean PPL(Q)/PPL(base)         :   1.059593 ±   0.001805
Mean PPL(Q)-PPL(base)         :   0.503318 ±   0.015365

====== KL divergence statistics ======
Mean    KLD:   0.104258 ±   0.000668
Maximum KLD:  18.479706
99.9%   KLD:   3.157249
99.0%   KLD:   0.989329
99.0%   KLD:   0.989329
Median  KLD:   0.042291
10.0%   KLD:   0.000107
 5.0%   KLD:   0.000012
 1.0%   KLD:   0.000000
Minimum KLD:  -0.000004

====== Token probability statistics ======
Mean    Δp: -1.473 ± 0.025 %
Maximum Δp: 99.114%
99.9%   Δp: 53.562%
99.0%   Δp: 24.617%
95.0%   Δp: 10.221%
90.0%   Δp:  4.865%
75.0%   Δp:  0.294%
Median  Δp: -0.017%
25.0%   Δp: -2.146%
10.0%   Δp: -9.929%
 5.0%   Δp: -17.075%
 1.0%   Δp: -37.839%
 0.1%   Δp: -77.867%
Minimum Δp: -99.916%
RMS Δp    :  9.937 ± 0.057 %
Same top p: 86.826 ± 0.088 %