Contributed to the yhyang201/sglang repository by implementing LoRA path support for chat APIs, enabling enhanced model customization through optional parameters in chat completion requests. This work established a foundation for broader LoRA integration in chat workflows, with careful attention to backward compatibility and API design. In the ROCm/pytorch repository, addressed a critical bug in the Fully Sharded Data Parallel (FSDP) module by introducing tensor flattening before writeback under NO_SHARD, resolving shape mismatch errors and improving distributed training reliability. Demonstrated expertise in Python, PyTorch, and distributed systems, focusing on robust backend development and deep learning model workflows.