====== Perplexity statistics ======
Mean PPL(Q)                   :   9.384561 ±   0.066615
Mean PPL(base)                :   9.366577 ±   0.066397
Cor(ln(PPL(Q)), ln(PPL(base))):  99.96%
Mean ln(PPL(Q)/PPL(base))     :   0.001918 ±   0.000206
Mean PPL(Q)/PPL(base)         :   1.001920 ±   0.000207
Mean PPL(Q)-PPL(base)         :   0.017984 ±   0.001945

====== KL divergence statistics ======
Mean    KLD:   0.001779 ±   0.000018
Maximum KLD:   1.977490
99.9%   KLD:   0.040032
99.0%   KLD:   0.012342
99.0%   KLD:   0.012342
Median  KLD:   0.001053
10.0%   KLD:   0.000028
 5.0%   KLD:   0.000005
 1.0%   KLD:  -0.000001
Minimum KLD:  -0.000152

====== Token probability statistics ======
Mean    Δp: -0.027 ± 0.003 %
Maximum Δp: 49.334%
99.9%   Δp:  7.267%
99.0%   Δp:  3.236%
95.0%   Δp:  1.482%
90.0%   Δp:  0.845%
75.0%   Δp:  0.164%
Median  Δp: -0.000%
25.0%   Δp: -0.202%
10.0%   Δp: -0.938%
 5.0%   Δp: -1.620%
 1.0%   Δp: -3.443%
 0.1%   Δp: -7.356%
Minimum Δp: -46.022%
RMS Δp    :  1.101 ± 0.012 %
Same top p: 97.794 ± 0.038 %