troubleshooting

Common problems

Fixes for the most common DeepLM Ops problems.

I cannot pull the image

Check these in order:

  1. You accepted the DeepLM GitHub organization invite.
  2. Your token is a classic token with the read:packages scope. Fine-grained tokens do not work with ghcr.io.
  3. You ran docker login ghcr.io -u <your-github-username> and pasted the token at the Password prompt.

See Create your GitHub token.

It never says "model ready"

The first start downloads the model once (2.55 GiB), so it can take a while. Watch progress with docker logs dlm-ops. Also check that Docker has at least 8 GB of memory: Give Docker enough memory.

Run docker restart dlm-ops, then docker logs dlm-ops for a new link.

The GPU is not used

  • On Apple silicon and other Arm machines the build is CPU-only, so --gpus all does nothing.
  • On amd64, put --gpus all right after docker run -d, not at the end of the command.
  • Linux needs the NVIDIA Container Toolkit. Windows needs Docker Desktop with WSL 2.

See Use an NVIDIA GPU.

The answers are about a sample cluster

You started the container with -e DLM_OPS_SLURM=replay, which turns on the demo cluster. Run the command without it to ask about your own machine. See Demo cluster vs your machine.