FBGEMM and FBGEMM_GPU Documentation Homepage

Welcome to the documentation page for the FBGEMM and FBGEMM_GPU libraries!

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 Caffe2 and PyTorch quantized operators on x86 machines.

FBGEMM_GPU (FBGEMM GPU Kernels Library) is a collection of high-performance PyTorch GPU operator libraries for training and inference. This library is built on top of FBGEMM and provides efficient table batched embedding bag, data layout transformation, and quantization support.

---

Table of Contents

General Info

• Contributing
• Documentation
• Contact Us
• License

FBGEMM Development

• Build Instructions

FBGEMM_GPU Development

• Build Instructions
• Installation Instructions
• Test Instructions

FBGEMM_GPU Overview

• Jagged Tensor Operators

FBGEMM C++ API

• Quantization Utilities

FBGEMM_GPU C++ API

• Sparse Data Operators
• Quantization Operators
• Pooled Embeddings Operators
• Table Batched Embedding Operators
• Jagged Tensor Operators
• CUDA Memory Operators
• Combine Input Operators
• Layout Transformation Operators
• Embedding Operators
• Experimental Operators

FBGEMM_GPU Python API

• Table Batched Embedding (TBE) Operators
• Jagged Tensor Operators