The technological evolution of Lagrange is not fragmented, but has a clear roadmap. From the initial Prover Network, to the SQL co-processor that addresses cross-chain data issues, and then to DeepProve-1, which directly targets AI inference verification, each step expands its capability as a 'universal verification layer'.
Currently, DeepProve has achieved complete verification of GPT-2 and is compatible with mainstream models such as LLaMA, Gemma, and Mistral, while also supporting arbitrary DAG structures, covering residuals, branches, and multiple inputs and outputs. This transforms it from a laboratory demo into a zkML system that can be practically implemented. In the future, Lagrange aims to expand to the verifiability of training and fine-tuning, allowing AI not only to prove the correctness of results but also to demonstrate the legitimacy and fairness of the training process.
This continuous iteration model ensures that Lagrange can always stay at the forefront of zkML technology. For developers and the ecosystem, this means it is not a 'one-time tool', but an evolving, long-term usable infrastructure.
