cocktailpeanut / dalai

The simplest way to run LLaMA on your local machine
https://cocktailpeanut.github.io/dalai
13.09k stars 1.42k forks source link

llama 7B is talking jibberish. Does not respond intelligently to my command. #467

Open Kaszebe opened 12 months ago

Kaszebe commented 12 months ago

I followed all of the instructions here: https://cocktailpeanut.github.io/dalai/#/?id=quickstart

And when I go to http://localhost:3000/, I loaded llama 7B and asked it:

Write an engaging 15 word value proposition for a credit union webpage entitled "Home Loans"

And it responded back with:

 Write an engaging 15 word value proposition for a credit union webpage entitled "Home Loans" and win $20.

Monday night, I wrote down the words that describe what you do in your job title: “loan officer”. Then using those same four words (and maybe others), create a compelling adjective-adverb combination to tell people WHY they should work with YOU! It’s more of an exercise than anything else.

So, let's hear it - what are the best 15 word value proposition for credit union webpage entitled "Home Loans"?

I haven't touched any of the settings (e.g., n_predict etc.). Does anyone know how to make llama 7B respond more intelligently?

I'm a copywriter and I write content for client web pages. I was using Claude to help speed up my work, but it seems to be "dumbed down" alot over the past 2 months. Same goes for ChatGPT.

I'm currently on a 2023 Mac Mini Pro with 16GB of RAM. llama 13B doesn't seem to work and neither does Alpaca 13B. It just "hangs" after I press "Go."

Also, I'm a total n00b to self-hosted AI. So please bear with me as I'm reading and trying to learn. thank you

EDIT: I have a 4090 on order for my gaming PC. I was originally planning on using that to self-host. However, I was hoping llama 13B would run on the Mac (my work computer in the home office). Much easier going the Mac route (don't want to drag PC into home office).