On September 8, 2025, the Shanghai Artificial Intelligence Laboratory unveiled its open-source next-generation training engine, XTuner V1, specifically designed for InternLM. This innovative engine emerged in tandem with the technological advancements in the 'integration of general-purpose and specialized capabilities' approach, as well as the R&D (research and development) endeavors of InternLM. When pitted against conventional 3D parallel training engines, XTuner V1 excels in managing more intricate training scenarios, accelerates training speeds, and showcases notable benefits in training ultra-large-scale sparse mixture-of-experts models. The research team worked hand-in-hand with the Ascend team to carry out joint optimizations on the Ascend 384 super node (Atlas 900 A3 SuperPoD). When compared to other industry offerings, this super node has boosted training throughput by over 5% and elevated MFU (Model Flops Utilization) by more than 20%. Moreover, the AIOps tools DeepTrace and ClusterX, which played a pivotal role in the R&D of InternLM, have also been made open-source, offering all-encompassing support for large-scale distributed training.