helicon-sequenceai 0.1.0

Lightweight ONNX Runtime predictor for Helicon SequenceAI models.

helicon-sequenceai

A small inference-only Python package for running ONNX models exported from the internal sequence-ai training repository.

What this package provides

• helicon.sequenceai.predict(...): a simple function that loads an ONNX model and runs inference.
• helicon.sequenceai.Predictor: a reusable predictor that caches the ONNX Runtime session.

What this package does NOT include

• The ONNX model file(s) themselves.
• Training pipelines (MLflow/Metaflow/Kedro), preprocessing/windowing, or the FastAPI UI.

You must provide the ONNX model path at runtime.

Install

pip install helicon-sequenceai

GPU (Linux x86_64 only, CUDA-enabled):

pip install "helicon-sequenceai[gpu]"

Note: the default install uses CPU-only onnxruntime. To use GPU, install the gpu extra and pass providers when calling:

preds = predict(features=X, model_path="model.onnx",
                providers=["CUDAExecutionProvider", "CPUExecutionProvider"])

Usage

from helicon.sequenceai import predict

preds = predict(features=[[1.0, 2.0, 3.0]], model_path="/path/to/model.onnx")

Development

With uv (recommended)

Create a virtualenv and install with dev extras:

uv venv
uv sync --group dev

Run tests:

uv run pytest

Build locally:

uv build

With pip

Install in editable mode with dev extras:

python -m pip install -e ".[dev]"
pytest