Developed and delivered a tensor parallelism configuration for Qwen3’s expert MLPs in the liguodongiot/transformers repository, focusing on enhancing scalability and performance within the model’s Mixture of Experts (MoE) path. The work centered on enabling distributed device execution, laying the foundation for future large-scale inference and training deployments. Using Python and leveraging deep learning and machine learning expertise, the implementation introduced a tp plan that prepares the codebase for robust performance benchmarking. No major bug fixes were addressed during this period, as the primary emphasis remained on feature completeness, code readiness, and supporting scalable multi-device model configuration.