Due to the sparse nature of Mixture-of-Experts (MoE) models, they are particularly suitable for hybrid CPU/GPU inference, especially in low-concurrency scenarios. This hybrid approach leverages both the large, cost-effective memory capacity of CPU/DRAM and the high bandwidth of GPU/VRAM. However, existing hybrid solutions remain bottlenecked by CPU computation limits and CPU-GPU synchronization overheads, severely restricting their ability to efficiently run state-of-the-art large MoE models, such as the 671B DeepSeek-V3/R1.

With the further improvement of industrialization, the problem of atmospheric pollution has become a major issue closely related to our human existence that is of common concern to scientists worldwide. The rapid development of remote sensing technology makes it possible to observe the atmosphere over a long period of time and on a large scale. In this study, we compared the Moderate-resolution Imaging Spectroradiometer (MODIS) aerosol product MCD19A2 with the NO2 retrieved by Sentinel-5 Precursor (Sentinel-5P). It is found that the data based on Sentinel-5P can clearly detect the local steel smelting enterprises while MODIS cannot, and the column density distribution trend has a better correlation with the steel output of the current month. If the spatial resolution of related products is improved in the future, this method of atmospheric observation can be better used for pollution monitoring.