MLCommons Releases MLPerf Training v6.1 Results, Spotlighting Recent Advances in AI Training
MLCommons released MLPerf Training v6.1 benchmark results featuring two new benchmarks: DLRMv4-HSTU for recommendation systems and Qwen3.5-397B for large-scale LLM post-training. The round achieved record participation with 153 systems from 24 organizations, including 37% cloud-based systems and the first submissions using NVIDIA Rubin-generation GPUs and Google TPUv7.