In March 2025, Baishihao developed a Torch Inference Profiling Feature for the ModelTC/lightllm repository, focusing on enhancing observability and performance optimization in model inference. Using Python and PyTorch, Baishihao integrated a torch_profile utility into the tppart_model_infer pipeline, enabling detailed profiling of both prefill and decode stages. This approach provided end-to-end visibility into forward-pass latency and resource usage, supporting data-driven optimization efforts. To ensure reliability, Baishihao added dedicated test coverage for the profiling tooling, emphasizing robust testing practices. The work demonstrated depth in performance profiling and model inference, addressing the need for actionable insights in deployment scenarios.