[coffeebeqn]

DeepSeek has some something pretty remarkable. It’s certainly not “just” fine-tuning a Llama or a GPT prompt. More of a order of magnitude optimization