InvokeAI
InvokeAI copied to clipboard
Docker compose changes
Summary
I'm not a coder so these changes definitely need to be reviewed before being accepted!
- I couldn't get run.sh to succeed, and it looked like this line was building the
--build-argsstatement but the following lines needed the values in those args as well. So I replaced that line with awhilethat builds the--build-argsstatement and assigns the vals to the vars. - I also had to change the service name from
invokeai-nvidiatoinvokeai-cuda. Despite the comment in the.envfile saying that the values should be "nvidia" or "rocm", "nvidia" doesn't work but "cuda" does. This line may be the reason that "cuda" is required. And since$GPU_DRIVERmust be set to "cuda" for an Nvidia GPU,$profileis set based on$GPU_DRIVER, and$service_nameis set based on$profile,$service_nameends up beinginvokeai-cuda(notinvokeai-nvidia). - I also added
compose.override.yamltorun.shin case a user has such a need. I use Traefik as proxy and middleware plugins ipwhitelist and traefik-forward-auth, so I usecompose.override.yamlso that I don't have to modifydocker-compose.yamland thus have git telling me I need to commit a change.
Related Issues / Discussions
NA
QA Instructions
NA
Merge Plan
NA
Checklist
- [x] The PR has a short but descriptive title, suitable for a changelog
- [ ] Tests added / updated (if applicable)
- [ ] Documentation added / updated (if applicable)