Developed a backward-compatible optional length check for the masked_select_jagged_1d function in the pytorch/FBGEMM repository, addressing the need for models with varying mask lengths to maintain consistent behavior during migrations. The solution was implemented in C++ and Python, leveraging GPU computing and PyTorch expertise. By introducing the check behind an opt-in configuration flag, the update preserved default performance and API compatibility, allowing teams to safely experiment without impacting existing workflows. Comprehensive usage guidance and impact notes were provided to support adoption, reflecting a careful approach to change management and a focus on reliability in production environments.