标签
一个交互式指南，解释了大语言模型中的推测解码和多令牌预测，涵盖了从拒绝采样到Qwen 3.6和Gemma 4中使用的MTP等技术，配有实时图表和滑块。
An interactive visual guide, 'how-llms-work', breaks down the entire lifecycle of Large Language Models based on Andrej Karpathy's lectures, covering data collection to post-training.