Fractional GPU in multi-GPU setting with Ray

I needed to load 16 model actors in a 2 GPU setting on a single node for inference. When I set @ray.init(num_gpus=2) and @ray.remote(num_gpus=0.125), this configuration is loading models only on one GPU and not on the second GPU. When one GPU is full no more actors are created saying that resources are full. How do I make sure both my GPUs are utilized when loading multiple actors in a fractional GPU setting?