I kept getting some error when I was setting up the completions api using the Custom LLM guide in the vapi docs, finally got it working after one hellova conversation with gpt4