NeuroShard AI

Decentralized AI Training Network

Quick Start

# Install
pip install nextaitech

# Run a node (you'll need a token from neuroshard.com)
neuroshard --token YOUR_TOKEN

What is NeuroShard?

NeuroShard is a decentralized network for training large language models. Contributors share their GPU/CPU power and earn NEURO tokens based on their contribution (Proof of Neural Work).

Key Features

• Swarm Architecture - Fault-tolerant, multipath routing for resilient training
• Async Training - DiLoCo-style gradient accumulation for 90%+ bandwidth reduction
• Earn NEURO - Get rewarded for contributing compute power
• Cryptographic Proofs - ECDSA-signed Proof of Neural Work
• Web Dashboard - Monitor your node at http://localhost:8000

Requirements

• Python 3.9+
• 4GB+ RAM (8GB+ recommended)
• GPU optional but recommended for training

GPU Support

For NVIDIA GPU support, install PyTorch with CUDA:

pip install torch --index-url https://download.pytorch.org/whl/cu118

Usage

Basic Node

neuroshard --token YOUR_TOKEN

With Custom Port

neuroshard --token YOUR_TOKEN --port 9000

Connect to Specific Tracker

neuroshard --token YOUR_TOKEN --tracker tracker.neuroshard.com:8080

Web Dashboard

Once running, open your browser to http://localhost:8000 to view:

• Node status and role
• Network statistics
• Training progress
• NEURO balance
• Resource usage

Architecture

NeuroShard uses a swarm-based architecture for maximum resilience:

• Dynamic Routing - If one node fails, work automatically routes to others
• Activation Buffering - GPUs never starve waiting for network
• DiLoCo Training - Local gradient accumulation reduces sync frequency by 90%+
• Speculative Checkpoints - Fast recovery from node failures

Links

• Website: https://neuroshard.com
• Documentation: https://docs.neuroshard.com
• Get a Token: https://neuroshard.com/register

License

MIT License - see LICENSE for details.