Manage a model profile

Manage a model profile

A GGUF is the model file on disk. A profile is the saved configuration LM Nexus uses when it starts that model.

You can keep the default profile or create variants with different runtime settings.

Before you start

You need at least one GGUF visible in Model Manager.

If you do not have one yet, follow Download a local model.

Edit an existing profile

1. Open Model Manager.
2. Select Profiles.
3. Find the profile you want to change.
4. Use the profile action with the tooltip Configure model.
5. Review the runtime settings you actually need to change.
6. Save the dialog.

Typical profile settings include runtime values such as context size, GPU layers, threads, batch settings, cache settings, sampling defaults, Jinja support, and advanced llama-server options.

Some advanced options, such as speculative decoding, appear only when the installed backend reports that it supports them.

Create a variant

Use the + action in the Profiles toolbar. Its tooltip is New Variant.

A variant starts from the model’s existing/default profile settings and becomes independent after creation. This is useful when you want, for example:

• one profile with a larger context;
• another profile tuned for lower VRAM use;
• a different speculative-decoding setup;
• separate experimental flags without disturbing the default profile.

Understand overrides

Profile values can override global Model Manager defaults.

The configuration UI marks overridden fields and provides reset controls so you can return a field to the inherited/default behavior instead of manually copying values back.

Expected result

After saving, the profile remains listed in Profiles with its own runnable configuration.

Creating a variant does not duplicate the GGUF itself. Multiple profiles can point at the same underlying model asset.

Before changing advanced settings

Use advanced runtime settings only when you know why you need them.

In particular:

• a VRAM estimate is guidance, not a guarantee that the model will start;
• custom llama-server flags can conflict with structured settings;
• speculative modes are enabled only when the installed backend/profile reports the required capability;
• Direct I/O is used only when the selected llama-server, host, and filesystem support it.

Next

Start or stop a model.