This is useful.Would you know of a resource that compares the current LLM leader GPT-4 with open weights models (like llama) ? So the idea being you could use the OpenAI API but hit a locally running open weights model instead of GPT-4. It's never clear to me what each model is capable of e.g. function calling.