Skip to content

README.md

Latest commit

 

History

History
46 lines (34 loc) · 2.28 KB

README.md

File metadata and controls

46 lines (34 loc) · 2.28 KB
Raw

Cloud TPU VM Auto-Provisioner & Environment Setup Automator

A robust Python utility script designed for ML researchers to automate the orchestration of Google Cloud TPU VMs, specifically tailored for Google TPU Research Cloud (TRC) users.

This script continuously hunts for available Spot TPU resources across multiple zones, automatically provisions the hardware, and securely initializes the software environment without failing from silent SSH command truncation.

Key Features

  • Multi-Zone Spot Hunting: Iterates through predefined zones and configurations to find available TPU slices (v4, v5e/v5lite, v6e) without manual monitoring.
  • State Verification: Automatically checks if a READY node already exists before attempting new creations to prevent redundant provisioning.
  • Robust Software Initialization: Fixes the common Windows/Linux gcloud ssh --command string escaping issue by dynamically packaging and transferring a native Bash script (tpu_setup.sh) via SCP.
  • Pre-aligned Version Mapping: Out-of-the-box support for strict version alignment between torch==2.9.0 and torch_xla==2.9.0 alongside numpy<2.0.0 constraints.

Prerequisites

This script requires the Google Cloud CLI (SDK) installed on your local machine.

  1. Install Google Cloud CLI: Follow the official installation guide for your operating system:

  2. Authenticate and Set Project: Run the following commands in your terminal to log in and set your default project:

    gcloud auth login
gcloud config set project YOUR_PROJECT_ID

Configuration

You can easily customize the zone targets and hardware specifications in the ZONES_CONFIG block within the script:

ZONES_CONFIG = [
    {"zone": "europe-west4-a", **TPU_V6E_64,     "spot": True, "name": "auto-tpu-v6e-64-eu"},
    {"zone": "us-central1-a",  **TPU_V5_LITE_64, "spot": True, "name": "auto-tpu-v5lite-64"},
]

Usage

Simply run the Python script on your local machine:

python tpu_auto_run_public.py

Acknowledgement

Special thanks to Google TPU Research Cloud (TRC) for empowering my research projects with computational resources.

License

This project is licensed under the MIT License.