I am running local models but noticed that Copilot didn't work with them correctly. To fix that I made a llama copilot extension that fixes some problems like Copilot throwing away thinking traces: https://marketplace.visualstudio.com/items?itemName=delft-so...

Getting pretty good results now with the Qwen models on my mac.