TorchAO is an easy-to-use quantization and optimization library built on native PyTorch. It provides quantization (INT8, INT4, FP8), sparsity, and low-bit optimizer support, and works out-of-the-box with torch.compile() and FSDP2. TorchAO is the official PyTorch quantization backend for vLLM, HuggingFace Transformers, and TorchTune.