本课程由 DeepLearning.AI 与 AMD 联合打造，Sharon Zhou 主讲，帮助你深入理解 Transformer 架构的 LLM 实际运作机制。你将学习模型如何逐个 token 生成文本、如何通过注意力机制决定哪些历史词汇影响下一个词的预测，以及量化等加速 GPU 推理的技术。课程包含互动可视化，让你亲手操作这些概念，建立直观直觉，并掌握诊断推理瓶颈、缓解模型幻觉以及运用 RAG 和思维链等实用技能。
Kog AI presents a technical talk on building a real-time inference stack using AMD Instinct GPUs, covering architecture, performance optimizations, and deployment strategies for AI inference workloads.
A professor building onboarding systems for LLMs identifies the need for a mental model that can be transmitted through language, drawing a parallel between teaching students and onboarding language models.