FBGEMM_GPU

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

FBGEMM_GPU Development

• Build Instructions
  • Set Up an Isolated Build Environment
  • Set Up for CPU-Only Build
  • Set Up for CUDA Build
  • Set Up for ROCm Build
  • Install the Build Tools
  • Install PyTorch
  • Other Pre-Build Setup
  • CPU-Only Build
  • CUDA Build
  • ROCm Build
  • Post-Build Checks (For Developers)
• Installation Instructions
  • Set Up CPU-Only Environment
  • Set Up CUDA Environment
  • Set Up ROCm Environment
  • Install Python Libraries
  • Install PyTorch
  • Install Triton
  • Install the FBGEMM_GPU Package
  • Post-Installation Checks
• Test Instructions
  • Set Uup the FBGEMM_GPU Test Environment
  • Running FBGEMM_GPU Tests
  • Running FBGEMM_GPU Benchmarks
• Feature Gates
  • Creating a Feature Gate
  • Enabling a Feature Gate

FBGEMM_GPU Overview

• Jagged Tensor Operators
  • High Level Overview
  • Jagged Tensor Format
  • Jagged Tensor Operations

FBGEMM Stable API

• FBGEMM_GPU Stable Python API
  • Stable APIs

FBGEMM_GPU C++ API

• Sparse Data Operators
  • CUDA Operators
  • CPU Operators
• Quantization Operators
  • CUDA Operators
  • CPU Operators
• Pooled Embeddings Operators
  • Merge Operators
  • Permutation Operators
• Table Batched Embedding Operators
• Jagged Tensor Operators
  • CUDA Operators
  • CPU Operators
• CUDA Memory Operators
• Combine Input Operators
• Layout Transformation Operators
  • CUDA Operators
  • CPU Operators
• Embedding Operators
  • CUDA Operators
  • CPU Operators
• SSD Embedding Operators
  • CUDA Operators
• Experimental Operators
  • Attention Operators
• Feature Gates (C++)

FBGEMM_GPU Python API

• Sparse Operators
  • Stable API
• Pooled Embedding Operators
  • Stable API
  • Other API
• Quantization Operators
  • Stable API
  • Other API
• Jagged Tensor Operators
  • Stable API
  • Other API
• Table Batched Embedding (TBE) Training Module
  • Stable API
  • Other API
• Table Batched Embedding (TBE) Inference Module
  • Stable API
  • Other API
• Pooled Embedding Modules
  • Stable API
  • Other API
• Feature Gates (Python)
  • Stable API