3 ms·Nvidia's new Llama-3.1 Nemotron Ultra outperforms DeepSeek R1 at half the size11 points by hochmartinez 2y agoHavoc 2y agoHalf the size is not a great metric when comparing a dense model against a MoE.dyl000 2y agollama has order of magnitude mode compute requirement than deepseek.