In the past year, zkML has remained in the academic and laboratory stage, with inadequate performance and limited compatibility.
However, in the engineering update in July, Lagrange announced the complete validation of GPT-2 and released the first production-ready DeepProve-1. This marks the transition of verifiable AI from a research concept to a reality that can be applied on a large scale.
The update of DeepProve-1 is not just an improvement in performance; more importantly, it is compatible with complex model structures: it supports graph-based computation DAG, covers the complete computational process of Transformers, and is compatible with the GGUF format, allowing for the direct loading of mainstream open-source models from the community.
This means that developers do not need to redesign the model structure to introduce verification into their existing AI workflows.
This 'backward compatibility' strategy ensures that Lagrange's technology can seamlessly integrate into existing systems in high-risk industries such as finance, healthcare, and defense.
Furthermore, the official announcement states that they will continue to provide transparent updates through monthly engineering updates, allowing the community to track the rapid evolution of zkML.
