https://howtonotcode.com/story/132-nvidia-groq-chatter-highlights-multi-backend-inference-planning