DeepLearning.AI Launches Official vLLM Course

A hands-on, short course covering the internal mechanisms of vLLM and production-level code examples has been launched on Andrew Ng's DeepLearning.AI platform.

Key topics covered include:

• KV Cache Bottleneck: Visualizing the impact of autoregressive decoding on VRAM bandwidth, and learning memory management through virtual block allocation
• Model Compression and FP8 Quantization: Hands-on practice implementing FP8 dynamic quantization while maintaining accuracy using LLM Compressor
• Production Profiling: Using GuideLLM to map a model's latency and requests-per-second (RPS) curves and perform stress testing

This course provides practical theory and optimization pipelines for engineers looking to optimize LLM serving using vLLM.