====== Perplexity statistics ======
Mean PPL(Q)                   :   7.275772 ±   0.045822
Mean PPL(base)                :   7.237090 ±   0.045539
Cor(ln(PPL(Q)), ln(PPL(base))):  99.93%
Mean ln(PPL(Q)/PPL(base))     :   0.005331 ±   0.000228
Mean PPL(Q)/PPL(base)         :   1.005345 ±   0.000229
Mean PPL(Q)-PPL(base)         :   0.038683 ±   0.001675

====== KL divergence statistics ======
Mean    KLD:   0.003219 ±   0.000027
Maximum KLD:   1.342281
99.9%   KLD:   0.106393
99.0%   KLD:   0.027442
99.0%   KLD:   0.027442
Median  KLD:   0.001673
10.0%   KLD:   0.000112
 5.0%   KLD:   0.000034
 1.0%   KLD:   0.000002
Minimum KLD:  -0.000176

====== Token probability statistics ======
Mean    Δp: -0.106 ± 0.004 %
Maximum Δp: 38.774%
99.9%   Δp:  9.164%
99.0%   Δp:  4.030%
95.0%   Δp:  1.858%
90.0%   Δp:  1.092%
75.0%   Δp:  0.234%
Median  Δp: -0.003%
25.0%   Δp: -0.371%
10.0%   Δp: -1.373%
 5.0%   Δp: -2.266%
 1.0%   Δp: -5.106%
 0.1%   Δp: -12.404%
Minimum Δp: -73.995%
RMS Δp    :  1.585 ± 0.020 %
Same top p: 97.349 ± 0.042 %