piterodml/Dperse

Dperse (or DSperse) is an academic-led project addressing the primary bottleneck of ZK-ML: the massive computational cost of generating proofs for large models. By implementing 'targeted verification,' it selectively proves only certain layers or sub-computations, which is a strategically sound approach for decentralized AI (DeAI) networks. However, with only 1 star and no forks after nearly 5 months, the project currently lacks any community momentum or production-ready tooling. It functions primarily as a reference implementation for the associated pre-print. In the competitive landscape, it faces significant pressure from more mature ZK-ML ecosystems like EZKL, Giza, and Modulus Labs, which are building broader infrastructure for model transpilation into ZK circuits. The moat is purely algorithmic/theoretical at this stage; without a robust library implementation that integrates with standard ML pipelines (like Hugging Face), it remains a niche academic experiment. Frontier labs are unlikely to compete here as they prioritize performance and scale over cryptographic auditability, leaving this niche to the Web3/DeAI sector.