cuvis-ai-schemas 0.7.0

Lightweight schema definitions for cuvis-ai ecosystem

cuvis-ai-schemas

Lightweight schema definitions for the cuvis-ai ecosystem.

Overview

cuvis-ai-schemas is a centralized, dependency-light package of schema definitions used across the cuvis-ai ecosystem. It enables type-safe communication between services without heavy runtime requirements.

Key points:

• Minimal deps (pydantic + pyyaml)
• Full Pydantic validation
• Optional extras for proto, torch, numpy, lightning, testing

Installation

uv add cuvis-ai-schemas
uv add "cuvis-ai-schemas[proto]"
uv add "cuvis-ai-schemas[full]"

Extras:

• proto: gRPC and protobuf support
• torch: PyTorch dtype handling (validation only)
• numpy: NumPy array support
• lightning: PyTorch Lightning training configs
• testing: Hypothesis strategies + round-trip assertions (cuvis_ai_schemas.testing)
• full: All features
• dev: Development dependencies

Usage

from cuvis_ai_schemas.pipeline import PipelineConfig, NodeConfig

pipeline = PipelineConfig(
    nodes=[NodeConfig(name="node_1", class_name="my_pkg.DataLoader", hparams={"batch_size": 32})],
    connections=[],
)

pipeline_json = pipeline.to_json()
pipeline = PipelineConfig.from_json(pipeline_json)

Development

uv sync --all-extras --dev
uv run pytest tests/ -v
uv run ruff check cuvis_ai_schemas/ tests/
uv run ruff format cuvis_ai_schemas/ tests/
uv run mypy cuvis_ai_schemas/

Contributing

Contributions are welcome. Please:

1. Ensure tests pass
2. Run ruff format and ruff check
3. Keep type hints and update docs as needed

License

Licensed under the Apache License 2.0. See LICENSE for details.