tract and rten

About tract

sonos/tract

Tiny, no-nonsense, self-contained, Tensorflow and ONNX inference

Implements graph-level optimization passes (constant folding, operator fusion, quantization-aware transformations) and supports symbolic dimensions for dynamic shapes, enabling efficient inference on resource-constrained embedded systems. Built in Rust with zero external dependencies, it provides both a standalone CLI and language bindings (Python, C) for framework integration. Handles ONNX (85%+ operator coverage), TensorFlow 1.x, and NNEF formats with a production-focused subset philosophy that excludes rarely-used features like tensor sequences in favor of maintainability and performance.

About rten

robertknight/rten

ONNX neural network inference engine