Install the app and check your Mac

Language Model Builder currently requires an Apple Silicon Mac and macOS 15 or later. This course was checked with version 1.2.2. Verify both details on the official download page because application requirements can change.

Choose About This Mac from the Apple menu. You want a chip line that starts with Apple M and a macOS version of 15 or later. If you see Intel in the chip line, this app will not run on your hardware today.

Download the notarized app from the official website or its GitHub releases. Do not download a copy from an unrelated mirror.

Keep enough free disk space for the app, datasets, checkpoints, and exported weights. A dataset download can be much larger than the final model file. I would leave at least 20 GB free for a first run with room for checkpoints.

Training uses unified memory shared by the CPU and GPU. Close memory-heavy applications before a long run, but record that choice in your notebook so timing comparisons stay fair.

Create a short environment record and fill it in before training:

Mac chip:
Unified memory:
macOS version:
App version:
Free disk space before the run:

Open the app once and create the project from the previous lesson. If macOS blocks the launch, confirm you used the official build before changing any security setting. Gatekeeper sometimes flags apps downloaded outside the App Store even when they are notarized.

The app keeps projects, datasets, checkpoints, prompts, and training results locally. Update checks and optional crash reporting have separate network behavior, so review the current privacy settings if the dataset is sensitive.

After creating the project, quit and reopen the app. Confirm the empty project remains available before committing hours to training.

What to observe: the project appears in the sidebar and shows that it has not been trained yet. If it disappears after reopen, fix that now. A lost project name is easier to debug before step 1 than after step 18,000.