Installation¶
Docker Compose (Recommended)¶
Prerequisites¶
- Docker and Docker Compose
- NVIDIA GPU with CUDA support (recommended for faster inference)
- At least 40GB disk space for LLM model storage
Setup¶
git clone https://github.com/dbcls/bsllmner-mk2.git
cd bsllmner-mk2
# Build and start containers
docker compose up -d --build
GPU Configuration¶
The compose.yml reserves all available NVIDIA GPUs by default. To restrict which GPUs are used, edit the CUDA_VISIBLE_DEVICES environment variable:
For Ollama performance tuning options, see Configuration - Ollama Performance Tuning.
uv (Local Development)¶
Prerequisites¶
- Python 3.10+
- uv
- Ollama server running locally or remotely
Setup¶
If the Ollama server is running on a different host, set the OLLAMA_HOST environment variable:
Verify Installation¶
For Docker Compose:
For uv (local):