Worked on the pytorch/ao repository to enhance the reliability of batch normalization folding in quantized deep learning models. Addressed a complex scenario where multiple convolutional layers share weights by updating the folding logic in the prepare_pt2e module, ensuring that quantization preserves both model accuracy and performance. Extended the test suite with a chunked batch normalization fusion test to validate correctness under shared-weight configurations, reducing the risk of regressions in production. Utilized Python and PyTorch, applying expertise in deep learning and quantization to stabilize the batch normalization folding path and improve robustness for shared-weight convolutional architectures.