Meta Andromeda: Powering the Next Generation of Ads Retrieval Through Model-System Co-Design
Meta has detailed the next generation of its personalized ads retrieval engine, Meta Andromeda, which utilizes a model-system-hardware co-design to evaluate over one billion ads per second on a single GPU server. By introducing a retrieval mixture-of-experts architecture and fusing model and system workloads into a single GPU kernel, the company achieved near-ranking-level performance at a fraction of previous inference-time FLOPs. The system operates on NVIDIA hardware and has delivered increased conversions on Facebook since its rollout. Open article →
- Meta Has A New Ad Ranking System Built On LLMs, Plus Upgrades To Andromeda And GEM
- Inside the Three New AI innovations in Ads Ranking Improving your Ad Performance
- Meta says its expanded processing capacity is driving ad results
- Sociable: Meta says its expanded processing capacity is driving ad results