Worked on enhancing the DecomposePower transformation in the tenstorrent/tt-tvm repository, expanding its capability to handle fractional exponents up to 0.75. This involved applying two square roots and a multiplication to support exponents beyond the previous single-square-root implementation, enabling more expressive numeric decomposition for model code generation. The work focused on Python and emphasized code refactoring and operator decomposition, with careful attention to maintainability and clear documentation. Improved error messaging was introduced to guide users when unsupported exponents are encountered, reducing debugging time and improving usability. The changes broadened the backend’s accuracy and robustness for non-integer exponent handling.