replicate / codespaces

Internal scripts and tools to make Codespaces work better

Replicate Codespaces scripts

This is a convenience script that installs Cog and nvidia-docker on GitHub's GPU Codespaces, so you can build run and push Cog models to Replicate.

To use it, run the following from the terminal in your Codespaces instance:

curl https://replicate.github.io/codespaces/scripts/install-cog.sh | bash

codespaces's Issues

unable to `cog run bash` on a fresh codespace

I ran this command on a fresh GPU codespace, starting from https://github.com/chenxwh/textual-inversion-codespace

curl https://replicate.github.io/codespaces/scripts/install-cog.sh | bash

Cog is installed:

@zeke ➜ /workspaces/textual-inversion-codespace (main) $ cog --version
cog version 0.4.2 (built 2022-08-23T14:57:26Z)

But cog run bash fails:

@zeke ➜ /workspaces/textual-inversion-codespace (main) $ cog run bash
⚠ Cog doesn't know if CUDA 11.3 is compatible with PyTorch 1.11.0 --extra-index-url=https://download.pytorch.org/whl/cu113. This might cause CUDA problems.
Building Docker image from environment in cog.yaml...
⚠ Cog doesn't know if CUDA 11.3 is compatible with torchvision 0.12.0 --extra-index-url=https://download.pytorch.org/whl/cu113. This might cause CUDA problems.
failed to dial gRPC: cannot connect to the Docker daemon. Is 'docker daemon' running on this host?: dial unix /var/run/docker.sock: connect: connection refused
ⅹ Failed to build Docker image: exit status 1

I'm not sure what's changed since yesterday that would make this crash and burn now.

Any ideas, @chenxwh and @andreasjansson?

errors with nvidia-container-toolkit and nvidia-docker2

Hey @andreasjansson and @chenxwh 👋🏼

I ran this in my Codespaces env:

curl https://storage.googleapis.com/replicate-public/codespaces/install-cog.sh | bash

And got this at the end of the output:

Errors were encountered while processing:
 nvidia-container-toolkit
 nvidia-docker2

Is that something I should be concerned about?

Full output: https://gist.github.com/zeke/8661da0fab3defd13911180fb807ff93

convert this to a Development Container™ template

The current process for getting Cog working in Codespaces is to run the curl one-liner in the README. That works, but it's a bit of a hack.

The paved-path way of boostrapping a Codespace with custom stuff is is to create a dev container template. See https://containers.dev/templates and https://containers.dev/implementors/templates/