[Feature] Inferencing using multiple backends

[snehargho]

Is there a plan to implement inferencing using multiple backends like llama.cpp? As in offloading a number of layers to GPU to control vram usage, gpu power draw, etc.