Basic concept of NNToolChain

SOPHONSDK

Sophgo’s original development tool kit for in-depth learning.

BM168x

Tensor Processing Unit released by Sophgo for in-depth learning field supports chips of BM1684 and BM1684X at present.

PCIE-Mode

A kind of product form. SDK is running on X86 platform. BM168x exists as the computing accelerator card of in-depth learning of PCIE.

SOC-Mode

A kind of product form. SDK runs independently on BM168x platform and supports interconnection with other equipment via Gigabit Ethernet.

TPU compiler

Refers to an optimizing compiler of deep neural network (DNN) developed against Sophon’s TPU processor. It can convert various DNNs defined by in-depth learning frame into instruction streams running on TPU.

BMNetC: Caffe model-oriented BMCompiler frontend

BMNetT: TensorFlow model-oriented BMCompiler frontend

BMNetM: MxNet model-oriented BMCompiler frontend

BMNetP: PyTorch model-oriented BMCompiler frontend

BMNetD: Darknet model-oriented BMCompiler frontend

BMNetO: ONNX model-oriented BMCompiler frontend

BMPaddle: PaddlePaddle model -oriented BMCompiler frontend