Worked on the ping1jing2/sglang repository to deliver the Longcat Flash N-gram Embedding feature, enhancing token representation for natural language processing tasks. This feature introduced n-gram embedding functionality within the Longcat Flash model, aiming to improve downstream NLP performance and prepare for future lightweight deployment with Longcat Flash Lite. The development process emphasized collaboration, with co-authored commits ensuring code quality and knowledge sharing. Leveraged Python, CUDA, and PyTorch to implement deep learning techniques and GPU programming for efficient model training and inference. No major bugs were addressed during this period, with the primary focus on feature development and business value.