neuralfn-python-sdk

What problem does it solve?

Provides a concise, code-first way to construct, validate, execute, and train neuron graphs using the neuralfn Python package so developers can avoid low-level boilerplate, inconsistent graph wiring, and error-prone serialization when building custom neural architectures.

Core Features & Use Cases

• Programmatic Graph Construction: Create NeuronDef instances, NeuronInstance placements, and Edge connections to define arbitrary directed neuron graphs.
• Multi-runtime Training & Execution: Support for scalar (function) execution, Torch module nodes, surrogate/evolutionary/hybrid training workflows, and PyTorch-backed TorchTrainer flows for large models.
• Serialization & Variant Libraries: Save/load graphs, round-trip to dict, resolve variant libraries, and export/import module state for checkpointing and deployment.
• Use Case: Quickly prototype a small neural graph for XOR with surrogate training, or assemble a GPT-style root graph using the provided presets and train via the Torch runtime.

Quick Start

Create a NeuronGraph with input, sigmoid, and output nodes, execute it with sample scalar inputs, then save the graph to a JSON file.