llama.cpp

Introduction

The llama.cpp module runs local model inference and a small proxy service using the unstable package set.

Requirements

The host must provide sufficient CPU, memory, and optional accelerator support for the selected model workload. Model downloads also require outbound access to the configured model source.

Persistence

The module persists /var/lib/llama-cpp, including downloaded model and Hugging Face cache data, through rootFs.persistDirectories.

Troubleshooting

Inspect llama-cpp-proxy.service; the llama-server process runs as its child rather than as a separate unit. Check /var/lib/llama-cpp and /var/lib/llama-cpp/hf-cache for model and cache state. Probe http://127.0.0.1:11434/_status to distinguish proxy failures from the internal server on 127.0.0.1:18080.

Options

llamaCpp.enable

Enable the llama.cpp proxy service.

Type: boolean

Default:

false

Declared by: