Complete source code for SFT and async GRPO reinforcement learning recipes on Qwen3 models using Microsoft Foundry, Ray, and SLIME — with a multi-turn retail environment and Streamlit dashboard. From Microsoft Build 2026. - View it on GitHub