Simulate Temperature and Top-P effects on token probability distribution
Temperaturecontrol softmaxOutputseparate " degree"。T→0whenseparate one-hot(totalisselectmosthighprobability)，T→∞when equal separate 。actualeffect：lowTemperaturefitcombinedneeddeterministic task(code、number )，highTemperaturefitcombined task(writework、head )。
Top-P probabilityfilter selecttoken。examplesuch asTop-P=0.9when，fromprobabilitymosthigh tokenStart add，untiltotaland to0.9，onlyfromThesetokenin 。thanTop-Kmore ： modelveryconfirmsetwhen selectfew，notconfirmsetwhen selectmultiple。
Top-K setprotected probabilitymosthigh K token。K=1etc in Decode，K=40iscommon defaultvalue。andTop-Pcan whenusing，get set。
allbaseinAlreadyGeneratetoken add ，but different：Frequency Penalty exit subnumberlinearincrease (exit 3sub 3 )，Presence Penaltyas long asexit throughthen add set (exit 1suband10sub same)。
Free online LLM parameter playground. Visually adjust Temperature, Top-P, Top-K, Frequency Penalty and more. Real-time preview of parameter effects. Pure frontend. Try now!
Free online LLM parameter playground. Visually adjust Temperature, Top-P, Top-K, Frequency Penalty and more. Real-time preview of parameter effects. Pure frontend. Try now!
Free online LLM parameter playground. Visually adjust Temperature, Top-P, Top-K, Frequency Penalty and more. Real-time preview of parameter effects. Pure frontend. Try now!
Enter your content in the input area, click the action button, and view the result. All processing is done locally in your browser.
notwill。AllParameterstunesectionandcanvisualizationallinBrowsetoolLocalcomplete，notSendAnyDatatoserver。