​
What’s New in 2.5?

  • DINOv2 Now Available: Train and fine-tune DINOv2, a powerful self-supervised vision model for high-quality image representations without labeled data.

  • Major Performance Improvements: Experience up to 77x speedup in checkpointing and weight initialization.

  • S3 Checkpointing: Save and manage model checkpoints using any S3-compatible storage for seamless remote training and distributed workflows.

  • Automatic Job Restart: Recover from intermittent failures automatically with checkpoint-based job restarts, ensuring uninterrupted training.

And many more updates! Read more in our release notes.

Getting Started

Setup and install your training environment.

Core Concepts

Learn more about our Wafer-Scale Cluster.

Model Zoo

Explore our comprehensive repo of deep learning models optimized for the Cerebras hardware.

Cs Torch API

Functions and data structures you can use to configure and execute PyTorch models on a Cerebras cluster.