====== Perplexity statistics ======
Mean PPL(Q)                   :   7.262551 ±   0.045671
Mean PPL(base)                :   7.237090 ±   0.045539
Cor(ln(PPL(Q)), ln(PPL(base))):  99.96%
Mean ln(PPL(Q)/PPL(base))     :   0.003512 ±   0.000175
Mean PPL(Q)/PPL(base)         :   1.003518 ±   0.000176
Mean PPL(Q)-PPL(base)         :   0.025461 ±   0.001278

====== KL divergence statistics ======
Mean    KLD:   0.001873 ±   0.000019
Maximum KLD:   1.001309
99.9%   KLD:   0.074453
99.0%   KLD:   0.018395
99.0%   KLD:   0.018395
Median  KLD:   0.000791
10.0%   KLD:   0.000068
 5.0%   KLD:   0.000022
 1.0%   KLD:   0.000001
Minimum KLD:  -0.000144

====== Token probability statistics ======
Mean    Δp: -0.125 ± 0.003 %
Maximum Δp: 35.516%
99.9%   Δp:  7.665%
99.0%   Δp:  2.917%
95.0%   Δp:  1.183%
90.0%   Δp:  0.651%
75.0%   Δp:  0.115%
Median  Δp: -0.008%
25.0%   Δp: -0.312%
10.0%   Δp: -1.020%
 5.0%   Δp: -1.651%
 1.0%   Δp: -3.901%
 0.1%   Δp: -10.195%
Minimum Δp: -44.522%
RMS Δp    :  1.209 ± 0.015 %
Same top p: 98.178 ± 0.035 %