CNN Agent Library on Wave DPU (MIMD)

DPU accelerated CNN Agent Library for inserence
  • Provides CNN building blocks optimized for the Wave DPU
Based on Google reference design (CBOW)
  • Conversion of floating point reference design to fixed point
  • Mapped complex software implementation to MIMD Wave DPU
Parameterized building blocks
  • Convolution +ReLU / PReLU + rounding
  • Pooling (max/avg)
  • VectorAdd, Scale, Batch-Norm, etc