Skip to content

chore: Create cuda runtime docker from scratch and drop NCCL - #1931

Open
fcollonval wants to merge 2 commits into
leejet:masterfrom
webscit:ft/slimer-docker
Open

chore: Create cuda runtime docker from scratch and drop NCCL#1931
fcollonval wants to merge 2 commits into
leejet:masterfrom
webscit:ft/slimer-docker

Conversation

@fcollonval

Copy link
Copy Markdown

Summary

I wanted to share a modification I did to the Dockerfile.cuda to make it slimmer by only bundling the tool in a scratch container.

Here is the reported sizes targeting a ARM device (namely Jetson Thor):

Image Disk usage Content size
sd-cpp:default (before) 5.56GB 2.28GB
sd-cpp:scratch (after) 1.59GB 670MB

It also disable NCCL (NVIDIA Collective Communication Library https://developer.nvidia.com/nccl) has this is not useful in the general case of local inference.

Related Issue / Discussion

None

Additional Information

One consequence is that the container is "empty" meaning there is no point to try to execute it interactively as there is not even a shell - only the final stable-diffusion.cpp binaries.

Checklist

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant