Modal Serverless Compute

Run Python and GPU workloads serverlessly with Modal. Covers defining functions and images in code, on-demand GPU containers, scale-to-zero, scheduled and web endpoints, volumes and secrets, and fan-out parallelism for batch inference and data jobs.

This skill helps you run compute on Modal. It defines container images and functions entirely in Python, provisions on-demand CPU and GPU containers with scale-to-zero, exposes web and scheduled endpoints, mounts volumes and secrets, and fans out massively parallel jobs for batch inference, model serving, and data processing without managing infrastructure.

When to use

Use when running Python or GPU workloads on Modal — code-defined images and functions, on-demand GPU containers, scale-to-zero, web/scheduled endpoints, volumes and secrets, or parallel batch jobs.

Examples

GPU inference endpoint

On-demand model serving

Define a Modal function that loads a model onto an A10G GPU, exposes it as a web endpoint, and scales to zero when idle

Parallel batch job

Fan out work

Use Modal to fan out embedding generation across 100 containers over a large dataset, writing results to a mounted volume