The Cortex project has made significant progress in the 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 the decentralized ecosystem. Cortex shared the update details via its official X account.
1/
Cortex Project Updates #127
Completed compilation of the MIPS instruction set for cvm runtime and deployed the test model.
Fixed issue with loading dynamic MIPS libraries and adapted to MLVM tensor/data types.
Changed memory format and modified model read format to… pic.twitter.com/DVMDJFHwOe
– Cortex Labs (@CTXCBlockchain) October 12, 2024
Completed Cortex’s MIPS instruction set and model modifications
The team successfully assembled the MIPS instruction set for the CVM runtime and implemented an extensive test model. In addition, they addressed issues with dynamic library loading, ensuring compatibility with tensor and data types for MLVM integration.
Significant progress has also been made in adapting the memory layout, allowing a more efficient read format that combines structure with weights and optimizes data processing for specific memory addresses.
Additionally, model inference within the CVM runtime now supports step functions, improving overall processing efficiency. The addition of the Unicorn debugging tool further streamlines the debugging process for developers.
Implementation and optimization of ZkRollup circuits
Cortex’s ZkRollup module saw significant improvements, especially in the areas of pre-compilation execution and static assertion removal. The circuit entry implementation has been completed and the team has deployed major deliverables to ensure queue parity. Code release requests have been sorted and deduplicated, and a new code decompressor has been added, which unpacks code directly into memory, optimizing execution speed.
The team also examined the Plonky3 compilation, which led to runtime optimizations. Research on 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 fixed, with several fixes providing better functionality. The team has introduced a configuration option that disables the FINDNODE liveness check, improving node performance. Added snapshots to the state database, and fixed bad tests being written to logs after completion.
