inference.cpp uses larod2. For most cases, larod2 is enough, since there is also no deprecation notice. However, there are some chips (in larod3 called devices) that are not available in larod2, like "axis-a8-dlpu-native", which accepts different model formats. This is useful to some teams.
inference.cpp uses larod2. For most cases, larod2 is enough, since there is also no deprecation notice. However, there are some chips (in larod3 called devices) that are not available in larod2, like "axis-a8-dlpu-native", which accepts different model formats. This is useful to some teams.