====== Perplexity statistics ======
Mean PPL(Q)                   :  10.039829 ±   0.072796
Mean PPL(base)                :  10.036835 ±   0.072696
Cor(ln(PPL(Q)), ln(PPL(base))):  99.99%
Mean ln(PPL(Q)/PPL(base))     :   0.000298 ±   0.000120
Mean PPL(Q)/PPL(base)         :   1.000298 ±   0.000120
Mean PPL(Q)-PPL(base)         :   0.002994 ±   0.001202

====== KL divergence statistics ======
Mean    KLD:   0.000162 ±   0.000001
Maximum KLD:   0.031653
99.9%   KLD:   0.002376
99.0%   KLD:   0.000973
99.0%   KLD:   0.000973
Median  KLD:   0.000112
10.0%   KLD:   0.000003
 5.0%   KLD:   0.000000
 1.0%   KLD:  -0.000004
Minimum KLD:  -0.000071

====== Token probability statistics ======
Mean    Δp: -0.006 ± 0.001 %
Maximum Δp:  7.425%
99.9%   Δp:  1.985%
99.0%   Δp:  0.955%
95.0%   Δp:  0.463%
90.0%   Δp:  0.270%
75.0%   Δp:  0.053%
Median  Δp: -0.000%
25.0%   Δp: -0.064%
10.0%   Δp: -0.293%
 5.0%   Δp: -0.489%
 1.0%   Δp: -0.973%
 0.1%   Δp: -1.929%
Minimum Δp: -9.811%
RMS Δp    :  0.319 ± 0.002 %
Same top p: 99.303 ± 0.022 %