What local model implements the coding instead of just suggesting changes to be made? I’ve tried DeepSeek R1, Llama3, CodeLlama, Mistral, Qwen2.5-Coder-14B. Nothing worked. Is it only with API access that coding implementing will work, so not just locally?
Hi,
did you set a proper context size in the .env file matching your hardware?
Other then that I would not recommend using local models. Even if you get something to work, and I testet a lot of models, the quality compared to cloud models is just bad and a lot more errors you need to iterate than using big ones.
There absolutely no sense for me, to use local models, as there are also free ones in the cloud like google Gemini, as well as very cheap ones, hosted by OpenRouter etc.