Granite Switch 4.1 gives chat, decisions and voice from one endpoint
The 8B and 30B models use about the same GPU memory as the base model.
The Granite-Switch-4.1-8B and Granite-Switch-4.1-30B models each include a decision adapter, a Guardian safety adapter and a speech recognition model. The adapters use the KV cache of the base model. Thus one endpoint can change between chat, decision and voice tasks.
Sources
Posted
