@Will S: The SoC is defined as RV32I, which is integer-only. There is no floating-point co-processor or implemented floating-point instruction set, so all floating-point will have to be done through software by linking to the appropriate library. I have done this before and it does work; I managed to compile TensorFlow Lite Micro for a RISC-V SoC. But I would have to dig down into my TensorFlow example to recall what I did to make it link to the proper libraries for floating-point software support.