dfalbel/gptneox-chat

comment stabilityai

go back to cpu

change runtime type to cuda.

remove shinyjs

make sure pkgconfig is installed

try a different dockerfile

some more organization

add spinning icons

Completely refactor the app for better shiny practices.

tokenize from the main session

add support fur gpu

Update title

Rm unused files.

Simplify to use task queue

new model

Use later for polling the process.

Invalidate more often.

Initialize the model just once.

Go back for cache invalidation.

Another try

Cache invalidation

Does it set a default token?

try again with changes from hfhub

MOdify docker for cache invalidation

try to get a better stack trace by. loading the model in the main process.

allow parallel sessions

Also reinstall minhub for dev versions

INstall hfhub

Fix msg

Actually no app.

Fix docker file

Use pre-built cpu binaries for faster installation.

maintain a session specific idinstead of global

Early return if generate is called and the session is not yet started.

Add handle condition for during model laoding.

Allow setting the model repo via env var.

use better model

Load model as part of the initial observer.

actually we don't need this.

delay session initialization.

Fix

reduce the number of messages upon failure.

maybe tmp?

another path

try another path

try another cache dir.

replace cache dir again

Improve error handling.

Better message when failure.

load magrittr for the pipe