Contributed to the mszep/tinygrad repository by developing a Tensor cloning API that enables the creation of independent tensor copies with separate memory buffers and correct gradient propagation. This feature was implemented in Python and involved careful handling of tensor operations to ensure gradients are managed accurately across various configurations. Additionally, addressed a stability issue in SDXL generation by ensuring the denoiser output is cloned and realized, preventing unintended in-place modifications during model inference. Expanded unit testing coverage to validate cloning and gradient scenarios, enhancing reliability and confidence in deep learning workflows. Work emphasized code refactoring, machine learning, and robust test practices.