search

epfml / disco

DISCO is a code-free and installation-free browser platform that allows any non-technical user to collaboratively train machine learning models without sharing any private data.
https://discolab.ai
Apache License 2.0
154 stars 26 forks source link

put ressources in repo instead of remotely #648

Open tharvik opened 8 months ago

tharvik commented 8 months ago

let’s also make sure we have the needed resources in our repos, if people want to deploy somewhere else in the future (cf @martinjaggi)

currently, our gcloud server is hosting some images, models, and datasets. we should not depend on it.

JulienVig commented 8 months ago

I think assets were removed from the repo in the past to make it more lightweight but we could rely on git-lfs rather than gcloud

tharvik commented 8 months ago

we could rely on git-lfs rather than gcloud

that's the way to go, and what I'm implementing for the wikitext dataset (used to test LLM models)

tharvik commented 6 months ago

Task have a DisplayInformation field that allow for an example image to be added. theses images are stored on Google storage. to have theses ressources locally (and to not immensly increase discojs-core bundle size), it will need to be provided via the server. thus implementing #647 makes sense to do before this issue,

tharvik commented 6 months ago

having users using git-lfs is non-trivial for general use. let's add a way to download theses from the server, and a button "use sample data" on the webapp.