====== Perplexity statistics ======
Mean PPL(Q)                   :  30.014168 ±   0.327772
Mean PPL(base)                :   9.042786 ±   0.071503
Cor(ln(PPL(Q)), ln(PPL(base))):  82.04%
Mean ln(PPL(Q)/PPL(base))     :   1.199702 ±   0.006332
Mean PPL(Q)/PPL(base)         :   3.319129 ±   0.021018
Mean PPL(Q)-PPL(base)         :  20.971382 ±   0.272199

====== KL divergence statistics ======
Mean    KLD:   1.004325 ±   0.003776
Maximum KLD:  22.233902
99.9%   KLD:  12.361748
99.0%   KLD:   7.606510
99.0%   KLD:   7.606510
Median  KLD:   0.570352
10.0%   KLD:   0.016693
 5.0%   KLD:   0.004282
 1.0%   KLD:   0.000351
Minimum KLD:  -0.000003

====== Token probability statistics ======
Mean    Δp: -7.018 ± 0.073 %
Maximum Δp: 99.437%
99.9%   Δp: 83.997%
99.0%   Δp: 60.615%
95.0%   Δp: 32.020%
90.0%   Δp: 16.340%
75.0%   Δp:  1.021%
Median  Δp: -0.417%
25.0%   Δp: -10.112%
10.0%   Δp: -43.816%
 5.0%   Δp: -71.991%
 1.0%   Δp: -98.342%
 0.1%   Δp: -99.994%
Minimum Δp: -100.000%
RMS Δp    : 28.772 ± 0.088 %
Same top p: 67.589 ± 0.122 %