Worked on enhancing distributed training performance in the NVIDIA-NeMo/Megatron-Bridge repository by updating the dispatcher for the GPT-OSS B200 V2 model. Focused on switching the default dispatcher to an alltoall communication pattern, which improved throughput and scalability for distributed systems. The implementation was completed in Python, emphasizing performance optimization while maintaining compatibility with existing APIs and minimizing the impact on the codebase. No critical bugs were reported during this period, and the integration was validated within Megatron-Bridge workflows. The work demonstrated a targeted approach to feature delivery, prioritizing efficient distributed training and robust system maintenance within a Python environment.