Developed a fused RMSNorm acceleration feature for the pytorch/pytorch repository, leveraging CUDA and Python to implement forward and backward kernels using the Quack-based cuteDSL framework. This work fully integrated with existing normalization kernels, focusing on both performance and maintainability. Comprehensive testing was conducted across OpInfo, test_nn, and test_ops suites, ensuring compatibility and parity with standard fused norm tests. The implementation relied on vendored Quack and included integration tests for cutedsl-filtered workflows. Collaboration with other contributors and adherence to PyTorch testing standards helped establish a robust foundation for future speedups in transformer-like models and improved code clarity.