Megatron Container Image Setting
Hi. I want to use 'nvcr.io/nvidia/pytorch:24.03-py3' image for using the megatron.
My start command is 'docker run --gpus all -it --rm -v /:/workspace/megatron -v /:/workspace/dataset -v /:/workspace/checkpoints nvcr.io/nvidia/pytorch:24.03-py3'
However, I have trouble with starting the pod.
How to fix the command?
5 Replies
You can't use
docker run in your start command, docker in docker is not allowed.
Also volume mount path handles your persistent storage volumes,
Just remove the start command completely and try change it to this instead:
The start command I think is the command to be executed in the docker container after it's started
Yes it is, it overrides
CMD.
It can't override ENTRYPOINT though.Thank you:)
Np