In September 2025, Srdjan Djordjevic enhanced the tenstorrent/tt-metal repository by implementing flexible tensor shape support within the Pybind module for ttnn.Tensor. He removed the previous hard-coded 4D shape constraint, introducing a vector-based shape representation at the Python-C++ interface. This approach allowed dynamic tensor shapes, improving integration with diverse models and pipelines while reducing shape-related errors. Srdjan’s work focused on bindings development using C++ and Python, resulting in a more usable and adaptable API for tensor construction. The depth of the change addressed a core limitation, enabling broader downstream compatibility without introducing new bugs during the release period.