Sessions & Technical Talks

Technical sessions and deep-dive talks on AI systems, machine learning, reinforcement learning, system design, and large-scale AI infrastructure. Each session breaks down complex concepts and connects research ideas with practical engineering.

LLM Inference & Serving Deep Dive: vLLM, SGLang, TensorRT-LLM & KV Cache Optimization — technical session thumbnail

LLM Inference & Serving Deep Dive: vLLM, SGLang, TensorRT-LLM & KV Cache Optimization

Oct 2026

How modern inference engines optimize memory, cut latency, and maximize throughput when serving LLMs at scale. Covers attention variants (MHA, MQA, GQA, MLA), KV cache sizing, prefill vs. decode, vLLM PagedAttention, SGLang RadixAttention and prefix caching, TensorRT-LLM, FlashInfer, continuous batching, Mooncake's disaggregated KV cache, Strata's hierarchical KV cache, prefill/decode disaggregation, quantization, and multi-GPU distributed serving.

LLM Inference · LLM Serving · KV Cache · vLLM · PagedAttention · SGLang · RadixAttention · TensorRT-LLM · FlashInfer · Mooncake · Prefill/Decode Disaggregation · GPU Optimization

Watch Session