AI Edge Quantizer
A post-training quantization toolkit for LiteRT and LiteRT-LM models, letting developers shrink vision and LLM models for deployment on mobile CPUs, GPUs and NPUs.
AI Edge Quantizer (AEQ) is a post-training quantization toolkit for LiteRT (formerly TensorFlow Lite) and LiteRT-LM models, aimed at optimizing vision models, LLMs and other GenAI pipelines for edge deployment. It supports selective quantization of specific layers by regex scope, mixed-precision schemes across layers, blockwise quantization, Hadamard transforms to suppress outlier activations, and the GPTQ and OCTAV optimization algorithms.
It also performs full static-range integer quantization, INT8 or INT16 activations with INT8 or INT4 weights, for mobile hardware, and includes built-in tensor-level distortion analysis such as MSE, SNR, cosine similarity and KL divergence compatible with Model Explorer. It ships as a command-line tool and a Python API, taking an unquantized .tflite or .litertlm model and a quantization recipe and producing a quantized model ready for edge deployment.