====== Perplexity statistics ======
Mean PPL(Q)                   :  23.519878 ±   0.223215
Mean PPL(base)                :  23.352232 ±   0.220841
Cor(ln(PPL(Q)), ln(PPL(base))):  99.96%
Mean ln(PPL(Q)/PPL(base))     :   0.007153 ±   0.000271
Mean PPL(Q)/PPL(base)         :   1.007179 ±   0.000273
Mean PPL(Q)-PPL(base)         :   0.167646 ±   0.006738

====== KL divergence statistics ======
Mean    KLD:   0.002318 ±   0.000013
Maximum KLD:   0.848010
99.9%   KLD:   0.043823
99.0%   KLD:   0.015801
99.0%   KLD:   0.015801
Median  KLD:   0.001166
10.0%   KLD:   0.000025
 5.0%   KLD:  -0.000001
 1.0%   KLD:  -0.000063
Minimum KLD:  -0.000401

====== Token probability statistics ======
Mean    Δp:  0.006 ± 0.003 %
Maximum Δp: 26.906%
99.9%   Δp:  7.636%
99.0%   Δp:  4.056%
95.0%   Δp:  1.919%
90.0%   Δp:  0.994%
75.0%   Δp:  0.127%
Median  Δp: -0.000%
25.0%   Δp: -0.119%
10.0%   Δp: -0.958%
 5.0%   Δp: -1.871%
 1.0%   Δp: -4.090%
 0.1%   Δp: -7.817%
Minimum Δp: -20.640%
RMS Δp    :  1.266 ± 0.008 %
Same top p: 97.778 ± 0.038 %

llama_perf_context_print:        load time =    3238.55 ms
llama_perf_context_print: prompt eval time = 2842839.16 ms / 304128 tokens (    9.35 ms per token,   106.98 tokens per second)
llama_perf_context_print:        eval time =       0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
llama_perf_context_print:       total time = 2975924.16 ms / 304129 tokens