post/llm-setting/

[giscus[bot]]

post/llm-setting/

LLM Settings Temperature Top-k Top-p Max Length Stop Sequences Frequency Penalty Presence Penalty Test Time Sampling Ref LLM Settings When designing prompts, We have few configs like Temperature, Top P, Max Length, Stop Sequences, Frequency Penalty, Presence Penalty. Below are the common setting and the explanation of them.
Temperature Temperature is the technique used to redistribute the probabilities of the possible values. Intuitively, it reduces the probabilities of common tokens, and as the result, increases the probabilities of rarer tokens.

https://v-vietlq.github.io/post/llm-setting/