FlowCompile

Installation

FlowCompile currently installs from source. The editable install exposes the flowcompile CLI and the flowcompile Python package used by the paper benchmark workflows.

Requirements

• Python 3.11 is the primary supported target.
• A virtual environment is strongly recommended.
• Full experiment runs may use heavyweight dependencies such as torch, vllm, litellm, and local model-serving packages.
• Hosted or OpenAI-compatible model endpoints can be used through the model configuration file without running local vLLM workers.

Install from Source

git clone https://github.com/UMass-Embodied-AGI/FlowCompile.git
cd FlowCompile

conda create -n flowcompile python=3.11
conda activate flowcompile

pip install -e .

Confirm the CLI imports:

flowcompile --help

The CLI has a flat config-driven interface with commands such as get-latency, prepare-data, profile, predict, test, runtime infer, and experiments correlation.

Install Docs Dependencies Only

The documentation build deliberately avoids the project runtime dependency set. For docs work, install the docs toolchain separately:

python -m pip install -r docs/requirements.txt

Build the Documentation Site

From the repository root:

python -m sphinx -n -W --keep-going -b html docs docs/_build/html

Or with the standard Sphinx make target:

make -C docs html

The HTML output is written to docs/_build/html.

This site is open source. Improve this page.