Skip to content

LINUX.md

Latest commit

 

History

History
68 lines (45 loc) · 2.61 KB

LINUX.md

File metadata and controls

68 lines (45 loc) · 2.61 KB
Raw

Experimental Linux Support

Linux support is experimental and currently supports Ubuntu and Fedora. InferBridge is still Windows-first, so start with CPU validation before spending time on GPU/NPU driver work.

Choose Your Distro

Common Commands

./setup.sh --minimal
./start_server.sh --mock
./start_server.sh --check-devices
./start_server.sh --model tinyllama-1.1b-chat-fp16 --device CPU

Use the platform-owned setup helpers for direct calls:

./setup/linux/install_deps.sh --minimal
./setup/linux/check_hardware.sh
./setup/linux/convert_model.sh --id tinyllama-1.1b-chat-fp16

The older setup/*.sh paths remain compatibility wrappers.

Linux desktop controller

The browser UI remains the primary Linux experience, but the desktop launcher can now be used from a source checkout:

./.venv/bin/python -m app.desktop_launcher --mock

If the desktop environment supports a compatible tray backend, InferBridge exposes the tray controls. If a tray backend is missing or fails to initialize, the launcher keeps the local server running instead of terminating it. The browser UI remains available at the local server URL.

Linux desktop data follows the XDG Base Directory convention:

  • $XDG_DATA_HOME/InferBridge when XDG_DATA_HOME is an absolute path.
  • ~/.local/share/InferBridge otherwise.
  • OV_LLM_DATA_DIR remains an explicit override.
  • Portable mode continues to use the launcher's local data/ directory.

Folder opening uses xdg-open with a gio open fallback. Clipboard actions support wl-copy on Wayland and xclip/xsel on X11 when those utilities are installed. Dialogs use zenity or kdialog when available and otherwise fall back to terminal/log output.

Hugging Face credentials

Linux setup never copies a Hugging Face CLI token into the repository .env file. For gated-model access, accept the model terms and export the token in the environment used to start InferBridge:

export HF_TOKEN=hf_...
./start_server.sh --model <model-id> --device CPU

This keeps the token user-controlled rather than duplicating it into the checkout.

Accelerator diagnostics

./setup/linux/check_hardware.sh reports:

  • OpenVINO-visible CPU/GPU/NPU devices and driver versions when available.
  • /dev/dri/renderD* access for Intel GPU execution.
  • /dev/accel/accel* access for Intel NPU execution.
  • render/video group membership and relevant kernel-module hints.

PCI visibility is only a hint. InferBridge only targets GPU or NPU when OpenVINO actually exposes that device.