Recently, at the MUSA open source technology salon, Yu Chao, head of the RLinF open source framework and an assistant professor at the Shenzhen Graduate School of International Studies at Tsinghua University, revealed that rLinF officially supports the operation of the Musa backend and has launched corresponding deployment documents; at the same time, the official CI has been connected to Mohr Thread MTT S5000 since version 0.3, and all code commits must complete automated verification based on domestic computing power before joining the main branch. At the same time, the two sides are jointly building a cloud-based intelligent framework based on domestic GPUs. The cloud uses MTT S5000 and the terminal side uses MTT E300, which covers the full range of training, inference, simulation and rendering scenarios. rLinF is the first “simulation and promotion integration” large-scale reinforcement learning framework for embodied intelligence. It was jointly released by Tsinghua University and Wuyuan Xinqiong. It was opened source in September 2025. The goal is to break through the fragmentation of the three links of rendering, training, and reasoning at the system level, so that developers can complete a complete closed loop from simulation environment interaction to strategy gradient updates on a unified platform.

Recently, at the MUSA open source technology salon, Yu Chao, head of the RLinF open source framework and an assistant professor at the Shenzhen Graduate School of International Studies at Tsinghua University, revealed that rLinF officially supports the operation of the Musa backend and has launched corresponding deployment documents; at the same time, the official CI has been connected to Mohr Thread MTT S5000 since version 0.3, and all code commits must complete automated verification based on domestic computing power before joining the main branch. At the same time, the two sides are jointly building a cloud-based intelligent framework based on domestic GPUs. The cloud uses MTT S5000 and the terminal side uses MTT E300, which covers the full range of training, inference, simulation and rendering scenarios. rLinF is the first “simulation and promotion integration” large-scale reinforcement learning framework for embodied intelligence. It was jointly released by Tsinghua University and Wuyuan Xinqiong. It was opened source in September 2025. The goal is to break through the fragmentation of the three links of rendering, training, and reasoning at the system level, so that developers can complete a complete closed loop from simulation environment interaction to strategy gradient updates on a unified platform.