Thread

I know the error (it should be llm = OpenAI(model="gpt-3.5-turbo"))

But OOC, why pydantic version do you have? pip show pydantic

aha

pretty sure it was tested with pydantic2

but I will fix it 🙂

(Thats why it wasn't noticed though)

Thank you Logan! I assume if I update to pydantic2, everything should be working

I think it should yes!

Or you can edit the installed template to have the code/fix I put above

Oh also, the llama-create app I've generated doesn't seem to be working. I give it a query and it just deletes the message entirely

that sounds sus. It worked just the other day lol

is there an error in the terminal anywhere?

no, thats the thing, its just silent

Maybe I can just run a simple CURL to the backend to see the problem or perhaps its a frontend issue

Hopefully, this video shows more context

there isn't any calls to the backend

Even the chrome console logs are empty? And the network logs?

Let me check on that

Hmm I think the console outlines CORS as an issue even though I remember seeing that yall allowed CORS by default

hmmm wait are you using firefox? Maybe try in chrome or edge 🤔

I think I figured it out actually

It's because im accessing it from my server perhaps

and edge reports this

ohhhh that might be why!

when I tested, I ran everything on the same machine

Weirdly though, I couldn't run create-llama on windows and thats why

classic windows. I used WSL

because the NEXT_*** probably was a linux env var

and it doesnt translate to windows

I forgot you can use WSL on VSCode on Windows

Got it working!

I just did some reverse ssh tunneling to my desktop

Question, so this does all types of docs and not just pdfs

like text,htmls or do I need loaders for seperate types?

yea, whatever files are in the data folder it will read automatically -- it handles nearly every file type

It's just using SimpleDirectoryReader() under the hood

I encourage you to dive into the template a bit to see where the magic happens 🙂 Good learning process!

One downside right now is that the index is loaded from disk on every query. Swapping in a hosted vector db would solve that, but it's also something we want to solve in the template soon 💪

Sweet, I imagine I can tweak that myself! Also, regarding the localhost issue, I assume I either need to expose the api as well or maybe do something with that?

yea likely expose the host/port and/or change the host/port its binding to