API Reference balanced_ternary_engine v#0.1.0-dev

Main API for converting pretrained models to balanced ternary format.

Per-architecture mixed-precision policies for common model families.

Simple benchmarking helpers for conversion and runtime execution.

Command-line interface for the BalancedTernaryEngine tool.

Exports and imports ternary models to/from .tbin binary format.

A neural network layer with ternary weights.

Packs ternary {-1, 0, +1} tensors into compact binary format.

End-to-end conversion: safetensors model → .tbin balanced ternary model.

Quantization-Aware Training layer with ternary weights.

Converts full-precision weight tensors to balanced ternary {-1, 0, +1}.

A dense weight matrix kept at half precision inside a converted model.

Minimal runtime for executing a converted .tbin model.

Pure-Elixir reader/writer for the HuggingFace .safetensors format.

Analyzes per-layer sensitivity to ternary quantization.

Aggregate statistics over a list of model layers.

Reusable validation helpers for tensors, configs, and model files.

Analyze a .tbin model.

Convert a pretrained model to balanced ternary format.

Run inference on a .tbin model.