Skip to content

Installation Overview

vllm-qaic can be installed in two modes. Each requires its own Python environment — they cannot coexist.

Installation Methods

Method Description Time
Docker (recommended for evaluation) Pre-built container from GHCR ~1 min
Scripted (recommended for development) ./scripts/install.sh aot | pyt ~5 min
Manual Step-by-step pip commands ~10 min
Wheel Pre-built wheels from SDK or CI ~3 min

Choose Your Mode

Best for production. Uses QEfficient to compile models into static QPCs.

Best for development. Uses torch_qaic for dynamic execution.