Is there a way to explicitly define which gpu should load the model?

[ub3rb3457]

in llama.cpp and ollama it's done using an environment variable (CUDA_VISIBLE_DEVICES)... is this functionality available using pieces copilot?