====== Perplexity statistics ======
Mean PPL(Q)                   :   9.379380 ±   0.066571
Mean PPL(base)                :   9.366577 ±   0.066397
Cor(ln(PPL(Q)), ln(PPL(base))):  99.98%
Mean ln(PPL(Q)/PPL(base))     :   0.001366 ±   0.000154
Mean PPL(Q)/PPL(base)         :   1.001367 ±   0.000154
Mean PPL(Q)-PPL(base)         :   0.012802 ±   0.001454

====== KL divergence statistics ======
Mean    KLD:   0.000692 ±   0.000012
Maximum KLD:   1.498512
99.9%   KLD:   0.024913
99.0%   KLD:   0.005779
99.0%   KLD:   0.005779
Median  KLD:   0.000331
10.0%   KLD:   0.000011
 5.0%   KLD:   0.000002
 1.0%   KLD:  -0.000002
Minimum KLD:  -0.000134

====== Token probability statistics ======
Mean    Δp: -0.028 ± 0.002 %
Maximum Δp: 21.822%
99.9%   Δp:  4.935%
99.0%   Δp:  1.949%
95.0%   Δp:  0.826%
90.0%   Δp:  0.458%
75.0%   Δp:  0.085%
Median  Δp: -0.000%
25.0%   Δp: -0.129%
10.0%   Δp: -0.558%
 5.0%   Δp: -0.952%
 1.0%   Δp: -2.125%
 0.1%   Δp: -5.152%
Minimum Δp: -39.341%
RMS Δp    :  0.706 ± 0.010 %
Same top p: 98.711 ± 0.029 %