Run MaxText

Choose your environment and orchestration method to run MaxText.

💻 Localhost / Single VM

Get started quickly on a single machine. Clone the repo, install dependencies, and run your first training job on a single TPU or GPU VM.

Via localhost or single-host VM

🎮 Single-host GPU

Run MaxText on single-host NVIDIA GPUs (e.g., A3 High/Mega). Includes Docker setup, NVIDIA Container Toolkit installation, and 1B/7B model training examples.

Via single-host GPU

🚀 At scale with Cluster Toolkit (GKE)

Deploy to Google Kubernetes Engine (GKE) using Cluster Toolkit’s gcluster CLI. Package and run multi-host JAX workloads with on-the-fly container builds.

At scale with Cluster Toolkit (gcluster)

🏗️ Legacy: At scale with XPK (GKE)

Deprecated. Retained for older deployments only. New GKE workloads should use Cluster Toolkit instead.

At scale with XPK

🌐 Multi-host via Pathways

Run large-scale JAX jobs on TPUs using Pathways. Supports batch and headless (interactive) workloads on GKE.

Via Pathways

🔌 Decoupled Mode

Run tests and local development without Google Cloud dependencies (no gcloud, GCS, or Vertex AI required).

Via Decoupled Mode (No Google Cloud Dependencies)

♻️ Elastic training (demo)

Demonstrate fault-tolerant training with Pathways on GKE: lose a TPU slice mid-run and recover in-process from the last checkpoint, no job restart.

Elastic training with Pathways