Switch to the latest llama.cpp

The local-AI scene moves fast. New models are released frequently, and newer llama.cpp versions often bring performance improvements, support for newly released models, and new features (better quantization, speculative decoding improvements, etc.). Keeping your build up to date is how you get the best speed and the widest model compatibility.

This guide shows you how to add a new llama.cpp build and use backend groups to swap it in for all your servers with a single click.

Step 1: Get a llama.cpp build

You need a llama-server binary for your OS and GPU. Two options:

• Download a prebuilt binary from the llama.cpp releases page. Pick the asset matching your OS and GPU type (CUDA for NVIDIA, ROCm for AMD, Vulkan for any GPU, or CPU).
• Compile your own — if you need specific flags or a custom build, see Recipes — Compile llama.cpp for your system.

Step 2: Add the backend

1. Go to the Backends page.
2. Click Add in the llama.cpp backends section.
3. Give it a name (e.g. llama-b5149-cuda), browse to the llama-server binary you downloaded, and add any default args if needed.
4. Click Save. warpdrv validates the binary and detects available GPU devices.

Step 3: Use a backend group

A backend group is a named collection of backends with one designated as active. Servers that reference the group always use the active backend — so you can swap which build they run without editing each server individually.

1. On the Backends page, expand the Backend Groups section.
2. Click Add to create a new group. Give it a name (e.g. my-cuda-builds).
3. Add the backends you want in this group (both the old and the new build).
4. Set the active backend — the one your servers should currently use.

Step 4: Point your servers at the group

For a server to benefit from group switching, it must reference the group rather than a direct backend:

1. Open the Launch Server dialog (Servers page → Launch Server, or edit an existing server).
2. In the Backend section, select the group instead of a specific backend.
3. Save and launch.

Any server set to use the group will always resolve to the group’s active backend.

Step 5: Swap to a new version

When you have a new llama.cpp build:

1. Add it as a backend (Step 2).
2. Add it to the group (Step 3).
3. On the group card, click the new backend to activate it.

A dialog appears with two options:

• Switch only — updates the active backend. Running servers continue with the old binary until you restart them manually.
• Switch & restart — updates the active backend and restarts all running servers in the group, so they immediately pick up the new build.

That’s it — all servers using that group now run the new llama.cpp version, with no per-server editing.

Next steps

• Recipes — Compile llama.cpp for your system — build a custom llama.cpp with specific flags.
• Use the model router — expose your models to external apps.