Comment by stupefy

What limits LLM inference accelerators? I heard about Groq (https://groq.com/) not sure how much it pushes away the problem.