wierd Answer from meta llama - 8b
Upon conducting a recent check, I came across the model of Meta Llama 8B. As part of my testing, I prompted the system with the question, "Hey, how are you doing today?" To my surprise, the response I received was quite unusual and unsettling. The answer seemed to be filled with a sense of fear and unease, which left me feeling quite perplexed and concerned. Despite this unexpected outcome.
Use the instruct model instead; you need more prompts to make the model productive.
Or help the girl anyway.
Wow, looks like your GPU needs an exorcism.
Try to use the instruction version and I recommend you to generate the text by using "repetition_penalty". I usually put it equal to 1.2. This way you will not have repetition problem.
I guess you haven't interacted much with the base models
this is a normal behavior of a base model, it just tries to randomly autocomplete your sentence. what you need is a chat model (also called instruct model often), which has the assistant personality
upd: oops, just realized i'm necroposting, never mind