Meta’s Mosaic uses a fleet of specialists for user embeddings
_reachsumit · x · 2026-07-28
Meta introduces Mosaic, a user-modeling platform that replaces a single shared user encoder with a fleet of specialists.
- The system uses four architecturally different specialist families: memorization-driven, dense-heavy, sequential-based, and CoTrain models.
- Two new techniques, MRM and CRL, are designed to maximize the marginal contribution of each added specialist.
- Meta also proposes CoEval and User Tower Zero-Out, a logging-free embedding evaluation framework meant to speed development while staying aligned with downstream accuracy.
- The serving stack is hybrid CPU/GPU and supports both online and offline serving depending on freshness, latency, and compute needs.
The paper reports consistent offline NE gains and online improvements.
More from Research
- Netflix details its production LLM judge: hundreds of thousands of recommendations scored weekly — omarsar0 · 2026-08-24
- Nature Comment: Provenance, not interpretability, grounds trust in autonomous science — gabepgomes · 2026-08-24
- New Architecture RHEA: Train 1B Model on 8GB VRAM — zemondza · 2026-08-24
- Trained two 16M-param models to do generative CAD with real physics — debreuil · 2026-08-24
- Claude model helps discover complex structure on S^6, solving 60-year-old math problem — Singularitarian · 2026-08-24
- Study: Agents read instructions/notes 60.5% of the time, rarely touch API docs — dair_ai · 2026-08-24