Install

Get pragmatiq running locally — Python 3.11+, a virtual environment, and optional extras.

pragmatiq is CPU-first: everything runs on CPU (slower but correct), and CUDA / flash-attn are accelerations, not requirements.

Requirements

git clone https://github.com/dynamiq-ai/pragmatiq.git
cd pragmatiq
python3.11 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

.[dev] adds the test/lint tooling (pytest, ruff, mypy). Run those dev commands from the virtual environment so they match CI.

Install only what a given workflow needs:

Extra	For
`.[gnn]`	the AML transfer-graph GraphSAGE ablation (`torch-geometric`)
`.[nemotron]`	the frozen-text-embedding variant (`transformers`)
`.[serve]`	ONNX export + Triton client
`.[gbdt]`	LightGBM as an optional probe head
`.[demo]`	the Streamlit demo
`.[wandb]` / `.[tb]`	experiment tracking

pip install -e ".[dev,gnn,serve]"   # combine as needed

pytest -q                 # the test suite
ruff check . && mypy pragmatiq
bash scripts/gates/gate_8.sh   # the CPU end-to-end pipeline smoke

Then run the Quickstart to see the whole pipeline end to end.