G.O.D Subnet

🚀 Welcome to the Gradients on Demand Subnet

Distributed intelligence for LLM and diffusion model training. Where the world's best AutoML minds compete.

🎯 Two Training Systems

1. Real-Time Serving

Miners compete to train models for Gradients.io customers who use our 4-click interface to fine-tune AI models.

2. Tournaments 🏆

Competitive events where validators execute miners' open-source training scripts on dedicated infrastructure.

• Duration: 4-7 days per tournament
• Frequency: New tournaments start 24 hours after the previous one ends
• Rewards: Significantly higher weight potential for top performers
• Open Source: Winning AutoML scripts are released when tournaments complete
• Tournament Overview
• Tournament Miner Guide

Setup Guides

• Real-Time Miner Setup
• Tournament Miner Guide
• Validator Setup Guide

Recommended Compute Requirements

Compute Requirements

Miner Advice

Miner Advice

Running evaluations on your own

You can re-evaluate existing tasks on your own machine. Or you can run non-submitted models to check if they are good. This works for tasks not older than 7 days.

Make sure to build the latest docker images before running the evaluation.

docker build -f dockerfiles/validator.dockerfile -t weightswandering/tuning_vali:latest .
docker build -f dockerfiles/validator-diffusion.dockerfile -t diagonalge/tuning_validator_diffusion:latest .

To see the available options, run:

python -m utils.run_evaluation --help

To re-evaluate a task, run:

python -m utils.run_evaluation --task_id <task_id>

To run a non-submitted model, run:

python -m utils.run_evaluation --task_id <task_id> --models <model_name>