During the month, contributed hardware-accelerated NZ weight format support for Ascend310P3 devices across the ggml-org/llama.cpp and Mintplex-Labs/whisper.cpp repositories. Developed C++ features enabling conditional conversion of tensor weights to the NZ format, leveraging low-level tensor operations and environment-variable driven configuration. Integrated backend logic with CANN to improve compatibility and performance for matrix multiplications on Ascend310P3 hardware, including helper utilities for weight format handling and tensor creation. The work focused on embedded systems and performance optimization, establishing efficient deployment pathways for llama models and related machine learning workloads without introducing new bugs, and maintaining code maintainability across both projects.