How much flexibility is there in the build commands?

with these AI packages for whatever reason, Poetry has been better at getting them running, so I was hoping I could use my custom docker image but we're still limited to openruntimes

can you guys allow us to use docker tags for our runtimes?

then we can publish our own runtimes muahaha that would be fun

so torch @Steven is a valid dependency in Poetry but in Python it fails

Poetry is the dependency management tool, right?

yeah so

it seems like Torch is broken

in this

alpine?

see I thought when you guys said any runtime it would be docker image based, so this is a let down, but it is what it is

i suggest commenting in https://github.com/appwrite/appwrite/issues/1037 to mention you need to use the python torch package but can't

okay, well my last comment is still there from before

I appreciate the Openruntime Upgrade, but please just change it so we can use our own docker image

I'm not a moron, but OpenRuntimes makes me feel like one with the way it's setup, I would rather spin up my own templates using FastAPI over flask and such for complete control

you mentioned pandas before and im pretty sure it's possible to get pandas working now

Yeah I can use Llama Index now

it just still didn't solve the alpine issue

The reason I love Appwrite is because it lets me do what I want how I want with my own self hosted instance, and you guys are amazing, but this is really limiting for me IMO. The fact that I can't even choose what image is pretty frustrating, because I am a competent developer, being able to at least give it a Dockerfile

would be ideal

the key is figuring out how to pass the vars to it

maybe sending a request to the docker image and then it builds it and passes the request to the build entrypoint

but how can I get pandas working?

if you want to get pandas working you could in theory run an extra command to install the missing dependencies, but I didn’t end up doing it

thanks. I noticed that appwrite cloud python runtime supports pandas/numpy, so I deployed the particular function needs pandas on appwrite cloud , and other functions on a self-host appwrite instance, then let them communicate with python SDK. It worked, although the experience is bumpy.