An open infrastructure to democratize and decentralize the development of superintelligence for humanity.
Psyche enables distributed training of transformer-based AI models over the internet, allowing collaboration between untrusted parties to train state-of-the-art machine learning models.
- Distributed Training: Train large AI models across thousands of nodes globally
- DisTrO Technology: Reduces GPU-to-GPU communication by 1000-10000x
- Blockchain Coordination: Uses Solana for transparent, censorship-resistant coordination
- Fault Tolerant: Handles node failures and dynamic participation
- Open Source: Fully transparent and community-driven development
- 40B parameters using Multi-head Latent Attention (MLA) architecture
- 20 trillion tokens of training data
- Largest distributed training ever conducted over the internet
- Accessible: Runs on single RTX 3090 GPU for inference
For detailed documentation on the Psyche project, please visit: 🔗 Official Documentation
- Language: Rust for core infrastructure
- Blockchain: Solana for coordination and incentives
- Training: DisTrO (Distributed Training Over-the-Internet)
- Consensus: Game theory and consensus mechanisms
- Hardware: Supports RTX 4090, A100, H100, and more
Current AI development is centralized in the hands of few large corporations. Psyche changes this by:
- Democratizing Access: Anyone can contribute computing resources
- Reducing Costs: Utilize idle hardware instead of expensive data centers
- Preventing Centralization: No single entity controls the models
- Enabling Innovation: Lower barriers for AI research and experimentation
This project is licensed under the Apache License 2.0 - see the LICENSE file for details.
Built by Nous Research - democratizing access to artificial intelligence through open source development and decentralized infrastructure.
"The future of AI should be open, decentralized, and accessible to everyone."
