Psyche: Nous Research's Decentralised AI Training Architecture

LinkNous Research TeamAnnouncement

Nous Research has unveiled Psyche, an open infrastructure designed to train large language models across distributed, heterogeneous hardware. By combining the DisTrO optimizer with Solana-based coordination, the architecture drastically reduces communication overhead to unlock idle global compute.

Key points

  1. DisTrO compression: Employs JPEG-inspired Discrete Cosine Transform (DCT) methods on decoupled momentum tensors to slash cross-node data transfer.
  2. 1-bit quantization: By transmitting only the sign (+1 or -1) and indices of top DCT momentum components, Psyche achieves an additional 3x bandwidth reduction.
  3. Overlapped training: Computations for subsequent steps begin while previous DisTrO updates are still transmitting, eliminating communication latency as a bottleneck at scale.
  4. Three core roles: An on-chain coordinator on Solana for state and random seed tracking, GPU clients that train, witness, and verify, and local or networked data providers.
  5. Epoch-driven lifecycle: Structured training epochs permit compute providers to onboard and offboard easily, reducing lock-in and compute idle costs.

Why it matters

Frontier AI model training has become heavily centralised due to massive hardware and bandwidth demands. Psyche offers a technical blueprint to coordinate heterogeneous, decentralized hardware, allowing community-driven training runs that rival traditional centralized data centres.

Analysis

Psyche directly addresses the two most critical barriers in decentralized training: bandwidth limits and trust. The combination of DCT compression and 1-bit quantization reduces gradient communication overhead by orders of magnitude, making consumer or standard internet connections viable. Meanwhile, shifting coordination, witness proofs (using Bloom filters), and random assignments onto a Solana smart contract provides a tamper-resistant consensus mechanism without a single point of failure.

Found in

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Similar in content, found by the AI search.

Let's talk about your project

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.