events
Faster AI: GPU Rooflines, Agent Learning & Homa’s Fast Transport Protocol
AI Performance Engineering Meetup (UK) organizza interventi online sull’utilizzo delle GPU, sul serving con vLLM, sull’apprendimento dalle tracce degli agenti e sul trasporto di piccoli messaggi di…

In inglese
- Starts
- 2026-10-19 18:00 UTC
- Organizer
- AI Performance Engineering Meetup (UK)
- Location
- Online event
The AI Performance Engineering Meetup (UK) hosts an online session with three technical talks on AI systems performance. Rajat Dandekar connects vLLM measurements on an H100 to GPU utilization and roofline modeling. Rustem Feyzkhanov discusses turning production agent traces into reproducible simulations, verifiable rewards, and model improvement through fine-tuning or reinforcement learning, and when to change prompts, tools, or the harness instead. John Ousterhout presents Homa, a transport protocol for small-message latency, and its open, workload-dependent relevance to AI. Chris Fregly and Antje Barth give introductions and updates.
Event website