Worked on the NVIDIA-NeMo/Megatron-Bridge repository to introduce quantization support for Megatron model weights prior to the resharding process. This approach reduced memory usage and accelerated deployment while maintaining compatibility with HuggingFace formats, enabling easier integration into existing machine learning workflows. Leveraged deep learning and model conversion expertise, utilizing Python and PyTorch quantization techniques to optimize model efficiency. The implementation focused on preserving interoperability with the HuggingFace ecosystem, broadening adoption and lowering infrastructure costs. Collaboration was emphasized through signed-off and co-authored commits, reflecting a team-oriented development process. No major bugs were addressed during this period, with efforts centered on new feature delivery.