I gave up on local AI model for 2 reasons.
Reason 1 is training time, which is simply too slow on any big training data, and small training data doesnt ever work good and the only useful result ends up being AI repeating from training data which is basically same as chatbot search program.
Reason 2 are AI hallucinations. I dont know how to prevent them, given that they happen to all AI models in existence.
So I decided to just cut the chase and make chatbot which simply repeats from data, and likewise has ability to search sentences which dont contain any word from input but are relevant to sentences which do.
My new default chatbot I use locally is made with this command to chatgpt:
"Give full code for chatbot which uses knowledge.txt file to generate response. It responds in a way that it finds sentence in knowledge.txt file with greatest amount of same words as input, and responds with that sentence. Then it finds sentence in knowledge.txt file which doesnt contain any words from input but contains greatest amount of words from the previous generated sentence, and responds with that sentence. It never generates same sentence twice even if same prompt is used. Make code as short as possible while keeping all functions."
