Worked on stabilizing the embedding quantization export workflow in the pytorch/torchchat repository, focusing on improving the reliability of exporting quantized models. Addressed a critical bug by introducing minimum and maximum quantization values to the 4-bit embedding export process, which enabled successful export for both 4-bit and 8-bit embeddings. This change reduced manual intervention and deployment delays, streamlining the quantized embedding workflow. The work involved deep debugging and instrumentation within the PyTorch ecosystem, leveraging skills in deep learning, model quantization, and Python. The targeted fix enhanced the stability of export commands, supporting smoother deployment of quantized models in production environments.