Skip to main content
Combines LFM2.5-Audio-1.5B in TTS and STT modes with LFM2-1.2B-Tool within a mockup of a car cockpit, letting the user control car functionalities by voice. All running locally in real-time.

​
Architecture

Llama.cpp is used for both models’ inference, with a custom runner for the audio model. The car cockpit (UI) is vanilla js+html+css, and the communication with the backend is through messages over websocket, like a widely simplified car CAN bus.

​
Quick start

Supported platformsThe following platforms are currently supported:
  • macos-arm64
  • ubuntu-arm64
  • ubuntu-x64
  • ubuntu-WSL2
1

Setup Python environment

make setup
2

Optional: Symlink llama-server

If llama-server is already in your PATH, symlink it instead of building:
ln -s $(which llama-server) llama-server
When building for ROCm, also install: sudo apt install -y libstdc++-14-dev
3

Prepare models

make LFM2.5-Audio-1.5B-GGUF LFM2-1.2B-Tool-GGUF
4

Launch demo

make -j2 audioserver serve
Building llama-server from sourceThe make -j2 audioserver serve step will build llama-server automatically if it is not already present. This requires cmake and a C++ toolchain. If the build fails, install the missing dependencies first:
PlatformCommand
macOSbrew install cmake (Xcode CLT required: xcode-select --install)
Linux / WSL2make install-deps
Then re-run make -j2 audioserver serve.

​
Source code

View the complete source code on GitHub.

Build docs developers (and LLMs) love

Get started for freeTalk to us