Thursday, November 12, 2026
1:30 PM – 3:00 PM PST
SGLang Summit Workshop by Nebius: Advanced GPU Inference Optimization
About this event
A 90-minute technical session from Nebius Token Factory's applied AI team on GPU inference optimization: sparse FP4 compression, GB300 serving with prefill/decode separation and speculative decoding, and agents that write CUDA kernels. Approval required, sold out with a waitlist.
A SGLang Summit workshop at Fort Mason presented by Nebius with LMSYS Org. Speakers from Nebius walk through next-generation hardware features and sparse FP4 compression, serving setups on GB300 systems such as prefill/decode separation, KV-cache locality and speculative decoding, and using AI agents to write CUDA kernels, with benchmark case studies tying hardware choices to model quality, throughput and latency.
The room seats 45 first come, first served, so registering does not guarantee a seat. Free but approval is required and an SGLang Summit pass (Full or Expo) is needed. It was sold out with a waitlist open when checked. 1:30 to 3:00 PM Pacific.
Organized by
Community events for SGLang, an open-source high-performance inference framework for large language models maintained by LMSYS Org.