Quickstart¶

Setup and data¶

This repository uses uv for package and project management. Start by syncing dependencies:

uv sync --dev

Then preprocess the data to create dataset splits and pair datasets under data/:

uv run python src/mlops_project/data.py

Train and serve¶

Train a model with the default config:

uv run python src/mlops_project/train.py

If you want to run on CPU or tweak a parameter, override the config on the CLI:

uv run python src/mlops_project/train.py meta.require_cuda=false train.batch_size=64 train.epochs=1

Run the embedding API with Uvicorn serving FastAPI:

MODEL_PATH="models/all-MiniLM-L6-v2-mnrl-100k-balanced" uv run uvicorn src.mlops_project.api:app \
  --host 0.0.0.0 --port 8000

Validate and iterate¶

Run the test suite with pytest:

uv run pytest tests/

Serve docs locally with MkDocs:

uv run mkdocs serve -f mkdocs.yaml