Welcome to Serverless GPU API’s documentation!

Serverless GPU API is a lightweight, easy-to-use library that simplifies launching multi-GPU and multi-node workloads on Databricks [Serverless GPU Compute](https://docs.databricks.com/aws/en/compute/serverless/gpu), directly from [Databricks notebooks](https://docs.databricks.com/aws/en/notebooks/).

It provides the same functionality as running launcher scripts such as torchrun train.py or accelerate launch train.py from the command line, but with seamless integration into the Databricks notebook environment.

Contents:

• Overview
  • What is Serverless GPU API?
  • Key Features
  • Architecture
  • Use Cases
  • Distributed Execution Details
  • Best Practices
  • Limitations
• Installation
  • Requirements
  • Installation Methods
• Quick Start Guide
  • Basic Usage
  • Pool Configuration
  • Ray Integration
  • Next Steps
• API Reference
  • serverless_gpu package
  • Core Modules

Getting Started

Serverless GPU API provides an easy way to run distributed computing workloads on Databricks. This documentation will help you get started with Serverless GPU API and understand its capabilities.

Features

• Easy-to-use interface for multi-GPU or multi-node workloads

• Integration with Databricks notebooks

• Comprehensive error handling and logging

Indices and tables

• Index

• Module Index

• Search Page