Meta· Llama / infrastructureModel update

From User Sequences to Scaling Laws: A Multi-Stage Architecture for Meta’s Ads Ranking

Meta introduces a multi-stage sequence model and dense tokenization with target-aware attention to scale sequence learning for ads ranking, achieving lifts in conversions and clicks.

From the source

Separating the sequence model into two complementary stages (upstream/offline user modeling and downstream/online ranking), enables model capacity to scale so that performance keeps improving without proportional increases in serving resources.

engineering.fb.com

Announced

05 Aug 2026 19:20 UTC