embedding-engine

Embedding backends (InsightFace/PyTorch + ONNX Runtime vs TensorRT). Use when optimizing embedding throughput or debugging numeric drift and fallback behavior.