events
Scaling Model Inference
Un intervento a Bengaluru che ricostruisce l’inferenza distribuita degli LLM a partire dai principi fondamentali, dai modelli roofline e dalla matematica dei Transformer, fino agli stack di serving e…

In inglese
- Starts
- 2026-11-18 13:00 UTC
- Ends
- 2026-11-18 14:00 UTC
- Organizer
- Lossfunk Event Calendar
- Location
- Bengaluru, India
A technical talk in Bengaluru on distributed LLM inference, organised by the Lossfunk Event Calendar. The speaker, Alhaad Gokhale, a Principal Engineer at Google DeepMind, builds the case from hardware rooflines and Transformer maths up to modern serving stacks, covering prefill/decode disaggregation, chunked prefill, prefix cache and speculative decoding, then turns to open research problems. The session draws on DeepMind's "How To Scale Your Model" and serving-systems literature, with two pre-reads listed. It is aimed at people working close to model serving and inference infrastructure.
Event website