Closed NavarrePratt closed 1 year ago
@NavarrePratt Build complete, success: https://github.com/coreweave/ml-containers/actions/runs/4185818831
Image: ghcr.io/coreweave/ml-containers/gpt-neox-mpi:npratt-fix-sd-inference-2cfbff3
@NavarrePratt Build complete, success: https://github.com/coreweave/ml-containers/actions/runs/4185818825
Image: ghcr.io/coreweave/ml-containers/sd-finetuner:npratt-fix-sd-inference-2cfbff3
@NavarrePratt Build complete, success: https://github.com/coreweave/ml-containers/actions/runs/4185818824
Image: ghcr.io/coreweave/ml-containers/sd-serializer:npratt-fix-sd-inference-2cfbff3
@NavarrePratt Build complete, success: https://github.com/coreweave/ml-containers/actions/runs/4185818827
Image: ghcr.io/coreweave/ml-containers/sd-inference:npratt-fix-sd-inference-2cfbff3
Upon further testing this doesn't fix the issue.
The build-args should be passed in as a new-line delimted list of strings, like in the
slurm.yml
workload. Without these changes, the action wasn't properly passing in the commit and would use the cached layer when building the docker image instead of pulling down the new changes fromkuberenetes-cloud
.Build using cached layers instead of pulling
kubernetes-cloud
Build with this fix properly pulling downkubernetes-cloud