Meta’s GRACE cuts generative ads decoder latency by 11×
_reachsumit · x · 2026-08-04
Meta presents GRACE, a serving system for generative ads retrieval designed for real-time production use.
The system adds decode-time eligibility filtering and kernel-level optimizations to cut decoder latency by 11×. The paper focuses on two production constraints: making generated ads respect targeting rules, and keeping latency and GPU cost low enough for high-volume, wide-beam decoding.
More from Companies & People
- MiniMax Exec Reviews Hailuo AI's Growth, Announces Push for Open Source — VoidAsuka · 2026-08-04
- Jensen Huang: Closed Models Are Actually Cheaper; Open Source Is About Control — rohanpaul_ai · 2026-08-04
- OpenAI appears to publicly call out Apple in a company post titled “Apple is getting this wrong” — apples_jimmy · 2026-08-04
- Numerai forms a team to build autonomous AI for quantitative research — AndrewCritchPhD · 2026-08-04
- Tencent’s Yuanbao search agent uses hybrid OPD to keep general skills — _reachsumit · 2026-08-04
- OpenAI says Apple emailed the wrong person in its lawsuit over file access — ChrisGPT · 2026-08-04