Install DeepLM Ops

DeepLM Ops runs on your own machine. On macOS and Ubuntu, one command installs everything it needs and starts it. On Windows, follow the Windows guide.

One command

Open a terminal on macOS or Ubuntu and run:

curl -fsSL https://deeplm.ai/dlm-ops-dev/install.sh | bash -s

It takes a few minutes. The first start downloads the model once (2.55 GiB). When it is ready, the installer opens DeepLM Ops in your browser.

What the installer does

1. Checks Docker, and installs it if it is missing. On macOS it installs Docker Desktop with Homebrew. If you do not have Homebrew, it asks you to install Docker Desktop yourself and run the command again. On Ubuntu it uses Docker's install script and adds you to the docker group. macOS or Ubuntu may ask for your password.
2. Starts Docker and waits for it to be ready.
3. Chooses how to run the model:
   • On an Apple silicon Mac with Docker Desktop 4.71 or later, it uses the Mac's GPU through Docker Model Runner. It turns Model Runner on and installs the host probe for you.
   • On Linux with an NVIDIA GPU, the NVIDIA driver and the NVIDIA Container Toolkit, it uses the GPU, as long as at least 4,608 MiB of GPU memory is free.
   • Otherwise it uses the CPU. That needs 8 GB of memory for Docker, and the installer stops with a message if Docker has less.
4. Downloads the image, ghcr.io/deeplm/dlm-ops-dev:0.2.0-rc.4. No account or token is needed.
5. Starts the container with the same locked-down settings as the manual command: a read-only file system, the page published on 127.0.0.1 only, and no extra Linux capabilities.
6. Waits for the model, then opens the sign-in link in your browser.

Options

Add options after bash -s --. For example, to use only this machine with no sample cluster:

curl -fsSL https://deeplm.ai/dlm-ops-dev/install.sh | bash -s -- --my-machine

The anonymous install report

The installer sends one small anonymous report when it starts, and one when it finishes or fails: the version, the kind of system, GPU or CPU, and whether it worked. It tells you before it begins, it never sends personal data, and it cannot slow down or break the install. Add --no-telemetry to turn it off. See Privacy and telemetry for the full list.

Run it again

The installer is safe to re-run. Each step checks its own state and skips work that is done, so the same command installs, upgrades, restarts and reopens DeepLM Ops. If you change an option, it replaces the container and keeps your data.

To start DeepLM Ops again later without the installer:

docker start dlm-ops && docker logs -f dlm-ops

Before you pipe a script into bash

You can read the script before you run it: install.sh is served as plain text. All of its work happens in functions, and the last line starts it, so a download that is cut off part way does nothing.

Other ways to install

If something goes wrong

• docker commands fail after the install on Ubuntu. Log out and back in, or run newgrp docker, so the new docker group applies.
• Port 8090 is in use. The installer stops and tells you. Stop whatever uses it and run the command again.
• Not enough memory for the CPU. Raise Docker's memory in Docker Desktop under Settings, then Resources. See Give Docker enough memory.
• Docker Desktop is too old on an Apple silicon Mac. Update it with brew upgrade --cask docker-desktop, then run the command again.
• The model is not ready after 30 minutes. Run docker logs dlm-ops to see why, and do not share any link it prints.

More fixes are in Common problems.