Ollama and num_gpu

[bchtrue]

Hello
Can you please add num_gpu ollama parameter to support by Zed?
I'm so disappointed! Tried Zed today with Local ollama. And to load model fully to GPU with Ollama I need to send to Ollama num_gpu 99 parameter, but it's not possible at Zed. Ollama is SLOW without this with Zed, because Ollama can not utilize all of my GPU memory and offload part of the model to CPU! (with 0 it will be possible to offload model fully to CPU if somebody want this)

for reference: ollama/ollama#6950 (comment)

[bchtrue]

....