Developed a TOSA backend legalization path for PyTorch’s bilinear upsampling within the llvm/torch-mlir repository, focusing on translating torch.aten.upsample_bilinear2d operations into TOSA-compatible tensor operations. The work involved designing a dedicated conversion pattern and implementing parameter-handling utilities to ensure accurate mapping of resizing parameters, enabling reliable end-to-end model conversion from PyTorch to TOSA-backed execution. Leveraging expertise in C++, MLIR, and tensor operations, the developer addressed compatibility challenges between PyTorch and TOSA, facilitating seamless integration for model deployment. The contribution centered on feature development, with an emphasis on correctness and maintainability in the codebase, without addressing bug fixes.