Modal Experiments

Run training jobs and experiments on Modal GPUs — from single-GPU fine-tuning to multi-node distributed training with RDMA. Write a proper Modal training app: bake your code into the image, mount volumes for data and checkpoints, add secrets, and set retries for fault tolerance.

For general Modal platform usage (app structure, function types, CLI, deployment), see the modal-basic-skills skill.

Quick Reference

Training App Pattern

A Modal training app is a single Python file:

import modal

app = modal.App("my-training")

data_volume = modal.Volume.from_name("training-data", create_if_missing=True)
ckpt_volume = modal.Volume.from_name("training-checkpoints", create_if_missing=True)