2026-05 monthly summary for AMD-AGI/Primus focused on business value and technical achievements. Delivered ML training workflow optimization and configuration management to boost training efficiency, reproducibility, and usability across environments. Highlights include updates to GLM5 and Minimax configurations for parameter optimization and modular execution flows, and a new streamlined script start_training_glm5_4layers_proxy.sh that simplifies startup and experimentation. Configuration management improvements enable reproducible runs across different environments, reducing setup time and risk. While no major bugs were reported this month for Primus, the work accelerates ML experimentation, improves training throughput, and strengthens cross-environment consistency. Technologies demonstrated include FP8/BF16 parameter optimization, shell scripting, configuration management, and deployment orchestration; collaboration is evidenced by a co-authored commit.