Worked on the jeejeelee/vllm repository to enhance type compatibility within the PyTorch ecosystem. Focused on stabilizing type handling by correcting the type annotation in the Fp8LinearMethod, expanding accepted input types from RoutedExperts to torch.nn.Module. This targeted bug fix reduced integration friction for downstream models and tooling, aligning with broader adoption and simplifying future maintenance. The work involved careful use of Python and type hinting to ensure seamless compatibility with existing PyTorch modules. Although the contribution was limited to a single bug fix over one month, it addressed a nuanced compatibility issue that supports ongoing development and integration efforts.