Yidao Information stated in the Investor Relations Activity Record Form that Yiling's AISSD heterogeneous inference technology performs low-level scheduling optimization for large MoE hybrid expert architecture models, reduces memory usage through dynamic expert residency strategies, and can be flexibly adjusted between memory and speed, so that endside devices with limited memory such as ordinary PCs and NAS can run 35B or larger models.

Yidao Information stated in the Investor Relations Activity Record Form that Yiling's AISSD heterogeneous inference technology performs low-level scheduling optimization for large MoE hybrid expert architecture models, reduces memory usage through dynamic expert residency strategies, and can be flexibly adjusted between memory and speed, so that endside devices with limited memory such as ordinary PCs and NAS can run 35B or larger models.