Implement if then else endif construction for the template language, do methods,
- repeat the same question multiple times, and then build a graph of answers.
- make a list of answers let the LLM select the best/ most appropriate one
- Eine Art Monte-Carlo-Tree search, bei der wir versuchen mehrere Antworten zu finden und dann entscheiden, welche die beste zu sein scheint.
The idea is to Let the model reason if done
- append another query part
QUERY
REASONING:
wait for EOS, then append with Summary, or someting else, to let the model refine the answer.
SUMMARY: