Worked on the tracel-ai/burn repository to deliver a quantization feature for BitNet b1.58, focusing on backend development and testing using Rust. Developed a ternary weight quantization calibration method based on absolute-mean symmetric range mapping, which enabled Q2S with PackedU32 storage and achieved substantial reductions in weight storage size. Enhanced the calibration abstraction by simplifying its interface and aligning it with the calibration-method model, improving code clarity and maintainability. Comprehensive test coverage was added for various calibration scenarios to ensure correctness, and code quality improvements were made through formatting and review-driven changes, resulting in a robust and efficient implementation.