MLPerf Training v6.1 results added a first post-training test, and the Qwen3.5 GRPO times from NVIDIA Vera Rubin (10.44 ...