vllm-cli — vLLM Model Serving CLI (Python)

Main

• Start in interactive mode (vllm-cli) when setting up GPUs/profiles, then switch to command-line mode for repeatable automation runs.

• Use built-in profiles and shortcuts to codify serving parameters; README shows serve --shortcut and hardware-optimized GPT-OSS profiles.

• Treat vLLM install as a separate compatibility step: README warns CUDA kernels must match PyTorch versions and vLLM-CLI won’t install vLLM by default.

Source-backed notes

• README documents Python 3.9+ support and multiple install options including pip install vllm-cli and pip install vllm-cli[vllm].
• README includes a basic usage snippet: vllm-cli serve --model openai/gpt-oss-20b.
• README notes vLLM binary compatibility concerns and recommends uv/conda-style installs for PyTorch/CUDA alignment.

FAQ

• Does vllm-cli install vLLM for me?: Not by default — README says vLLM-CLI will not install vLLM or PyTorch unless you use the extra.
• What is the first serving command to try?: README shows vllm-cli serve --model openai/gpt-oss-20b as a basic example.
• Why does install matter?: README warns vLLM uses pre-compiled CUDA kernels that must match your PyTorch version.