TMTPOST -- The team led by Professor Zhai Jidong from Tsinghua University's Institute of High Performance Computing, in collaboration with Tsinghua-affiliated tech startup Qingcheng Jizhi, announced the open-sourcing of their large model inference engine, "Chitu." This engine marks a significant breakthrough as it is the first to natively support FP8 precision models on non-NVIDIA Hopper architecture GPUs and various domestic chips. The newly open-sourced version specifically addresses a critical market demand: enabling the lossless and efficient deployment of the DeepSeek-R1 FP8 precision model on existing NVIDIA GPUs. The team also revealed that versions optimized for multiple domestic chips will be released in the near future. This development represents a major step forward in enhancing the accessibility and efficiency of large model inference, particularly for domestic hardware ecosystems, and underscores the growing capabilities of China's homegrown AI technologies.