Over a two-month period, contributed to the sglang repositories by building features that advanced distributed deep learning infrastructure and multimodal inference. Developed multimodal offline batch inference for visual language models in fzyzcjy/sglang, extending the engine to process both image and text data using Python and PyTorch. Added RDMA support in Docker environments to enable high-performance inter-node communication and introduced configurable distributed initialization timeouts to improve reliability. Enhanced documentation for quantized model usage and caching strategies. In bytedance-iaas/sglang, implemented tied-weights support for Qwen pipeline parallelism, ensuring accuracy parity and scalability in model-parallel deployments through robust testing and configuration management.