====== Perplexity statistics ======
Mean PPL(Q)                   :  24.304562 ±   0.233045
Mean PPL(base)                :  23.352232 ±   0.220841
Cor(ln(PPL(Q)), ln(PPL(base))):  99.74%
Mean ln(PPL(Q)/PPL(base))     :   0.039971 ±   0.000693
Mean PPL(Q)/PPL(base)         :   1.040781 ±   0.000721
Mean PPL(Q)-PPL(base)         :   0.952329 ±   0.020290

====== KL divergence statistics ======
Mean    KLD:   0.019001 ±   0.000084
Maximum KLD:   2.667997
99.9%   KLD:   0.307100
99.0%   KLD:   0.131595
99.0%   KLD:   0.131595
Median  KLD:   0.009005
10.0%   KLD:   0.000235
 5.0%   KLD:   0.000046
 1.0%   KLD:  -0.000033
Minimum KLD:  -0.000534

====== Token probability statistics ======
Mean    Δp: -0.033 ± 0.009 %
Maximum Δp: 53.497%
99.9%   Δp: 21.400%
99.0%   Δp: 11.091%
95.0%   Δp:  4.986%
90.0%   Δp:  2.607%
75.0%   Δp:  0.323%
Median  Δp: -0.000%
25.0%   Δp: -0.333%
10.0%   Δp: -2.621%
 5.0%   Δp: -5.142%
 1.0%   Δp: -11.765%
 0.1%   Δp: -23.329%
Minimum Δp: -88.396%
RMS Δp    :  3.498 ± 0.021 %
Same top p: 93.797 ± 0.062 %

llama_perf_context_print:        load time =  128672.98 ms
llama_perf_context_print: prompt eval time = 2206637.86 ms / 304128 tokens (    7.26 ms per token,   137.82 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 = 2358244.53 ms / 304129 tokens