The S100 and S100P have different memory capacities, and memory usage varies across different models. The base software for S100/S100P provides the hb_switch_ion.sh script to allocate memory space available for models. This script mainly supports the following two usage modes.
It is recommended to use balanced mode when running models of 3B or smaller. If you encounter errors while running a 7B model, please switch to bpu_first mode.
In bpu_first mode, models can utilize the maximum available memory space, making it suitable for running 7B models.