Gonka AI Compute offers a decentralized infrastructure for AI model training and inference, aiming to disrupt costly, centralized cloud providers. It targets AI developers needing scalable compute and hardware providers seeking to monetize their resources. The platform's core benefit is providing a cost-effective, efficient, and AI-aligned alternative for demanding computational workloads.

How It Works

Gonka introduces "Proof of Work 2.0," a novel consensus mechanism designed to allocate nearly 100% of computational resources directly to AI tasks, minimizing blockchain-specific overhead. Participants engage in time-bound "Races" using transformer-based models, aligning computation with AI workloads. A node's voting weight is determined by its successful computations, enabling a "one-computational-power-one-vote" governance and task allocation principle. Randomized Task Verification probabilistically ensures task integrity with significantly reduced overhead (1-10% of tasks), while Geo-Distributed Training leverages periodic synchronization for efficient, scalable decentralized training.

Quick Start & Requirements

To set up a local development environment:

1. Clone the repository: git clone https://github.com/gonka-ai/gonka.git
2. Build components and run tests: make local-build followed by make run-tests.

• Prerequisites: Git CLI, Go 1.22.8, Docker Desktop (4.37+), Make, Java 19+. NVIDIA CUDA is implied for MLNode execution.
• Resource Footprint: Local testing involves running a Docker cluster and can take considerable time.
• Documentation: Official documentation is available at https://gonka.ai/introduction/. Community discussions can be joined via Discord.

Highlighted Details

• Proof of Work 2.0 prioritizes AI workload computation over traditional blockchain security.
• "Races" utilize transformer models, directly aligning network activity with AI development needs.
• Voting weight and task allocation are directly proportional to proven computational contribution.
• Randomized Task Verification drastically reduces overhead while maintaining trust.
• Geo-Distributed Training enables scalable, decentralized model training with minimal communication overhead.

Maintenance & Community

The project maintains a Discord server for real-time discussions, updates, and support. Specific details on core contributors, sponsorships, or a public roadmap are not detailed in the provided README.

Licensing & Compatibility

The repository's license is not explicitly stated in the provided README. Compatibility for commercial use or closed-source linking is not specified. Model licenses are available via a separate link.

Limitations & Caveats

Diagram 1 is marked as "[Work in progress]". The documentation primarily focuses on local setup and testnet participation, suggesting the platform may still be under active development towards full production readiness.