• Quantization within TVM - Automatic Quantization • TVM stack ingests a FP32 graph and a small dataset • Finds suitable quantization scale • Produces a quantized graph • Compiling Pre-quantized models