Liquan Feng focused on stabilizing distributed deep learning workflows by addressing critical bugs in two major repositories. In kaiyux/TensorRT-LLM, he improved the WeightOnlyQuantRowLinear module by adding missing expert-mode logic, ensuring robust handling of distributed tensor processing during quantized inference. Later, in JustinTong0323/sglang, he resolved argument-passing issues in the fused MoE Triton benchmark, eliminating runtime errors and enabling reliable benchmarking. His work relied on Python and deep learning frameworks, with an emphasis on model quantization, benchmarking, and performance optimization. Across both projects, Liquan delivered targeted, high-impact fixes that improved reliability and maintainability in complex distributed systems.