FBGEMM

FBGEMM (Facebook GEneral Matrix Multiplication) is a low-precision, high-performance matrix-matrix multiplications and convolution library for server-side inference. This library is used as a backend of PyTorch quantized operators on x86 machines.

FBGEMM Development

• Build Instructions
  • FBGEMM Requirements
  • Set Up an Isolated Build Environment
  • Install the Build Tools
  • Build the FBGEMM Library

FBGEMM C++ API

• Quantization Utilities
  • Reference Implementation Methods
  • AVX-2 Implementation Methods
  • AVX-512 Implementation Methods
• TBE CPU Autovectorization
  • FP8/16/32 Autovec Implementation Methods