Cortex Unveils Key Updates on MIPS Compilation and ZkRollup Enhancements

The Cortex project has made significant strides in its recent update, focusing on MIPS instruction set implementation, ZkRollup circuit development, and Cortex Full Node optimizations. These developments further enhance the platform’s machine learning and blockchain performance, driving progress in its decentralized ecosystem. Cortex shared the update details through its official X account.

Cortex’s MIPS Instruction Set and Model Adjustments Completed

The team has successfully compiled the MIPS instruction set for the cvm-runtime and implemented a comprehensive test model. Additionally, they addressed issues with dynamic library loading, ensuring compatibility with tensor and data types for MLVM integration.

Significant progress was also made in memory layout modification, allowing for a more efficient reading format that combines structure with weights and optimizes data handling for specific memory addresses.

Furthermore, the model inference within the cvm-runtime now supports step functions, enhancing overall processing efficiency. The addition of the Unicorn debug tool further streamlines the debugging process for developers.

ZkRollup Circuit Implementation and Optimization

Cortex’s ZkRollup module saw key advancements, particularly in pre-compilation execution and removal of static assertions. Circuit input implementation is complete, and the team leveraged grand products to ensure queue equality. Code decommit requests were sorted and deduplicated, and a new code decompressor was added, which unpacks code directly into memory, optimizing execution speed.

The team also explored Plonky3 compilation, leading to runtime optimizations. Research into FAST-VM and naive VM execution processes helped refine the VM operational structure, improving the efficiency of the overall Cortex blockchain ecosystem.

In the Cortex Full Node section, minor issues in the state database were resolved, with several fixes ensuring better functionality. The team introduced a configuration option that disables the FINDNODE liveness check, improving node performance. Snapshots were added to the state database, and flaky tests writing to logs post-completion were fixed.