Slow Evaluation Speed

[Nabokov86]

Evaluation speed is significantly slower compared to llama-server alone. I've included performance metrics below for comparison.

llama-server with Mikupad:

prompt eval time = 24.59 ms / 1 tokens ( 24.59 ms per token, 40.67 tokens per second)
eval time = 1309.86 ms / 30 tokens ( 43.66 ms per token, 22.90 tokens per second)
total time = 1334.45 ms / 31 tokens

llama-server:

prompt eval time = 88.61 ms / 19 tokens ( 4.66 ms per token, 214.42 tokens per second)
eval time = 523.65 ms / 23 tokens ( 22.77 ms per token, 43.92 tokens per second)
total time = 612.27 ms / 42 tokens

llama-cli

llama_perf_context_print: prompt eval time = 103.19 ms / 12 tokens ( 8.60 ms per token, 116.29 tokens per second)
llama_perf_context_print: eval time = 1685.34 ms / 74 runs ( 22.77 ms per token, 43.91 tokens per second)

It seems Mikupad is causing a considerable slowdown. I'm wondering if this is due to how it handles token probabilities. Is it possible to disable this feature to potentially improve performance?