Meta’s Generative Ads Recommendation Model (GEM), the foundation model behind ads recommendations across Instagram and Facebook, now trains at LLM scale on several thousand of the latest-generation GPUs. The company achieved a doubling of end-to-end (E2E) training efficiency to 20–25% Model FLOPs Utilization (MFU) while scaling training FLOPs 4x in 12 months, by co-designing kernels, precision, parallelism, networking, and memory together. GEM combines a hybrid architecture with trillions of sparse embedding parameters and billions of dense parameters, trained on ad content and user engagement data with sequence features (e.g., user activity history) and non-sequence features (e.g., user location, ad creative representation). Its unique training challenges stem from the interplay between this hybrid architecture and recommendations-domain data properties, which differ from typical LLM workloads. AI infrastructure optimized for LLM training does not directly transfer, requiring significant innovation and hardware/software co-design. Compute efficiency was achieved through a customized recommendation kernel library—including Jagged Flash Attention (JFA), Generalized Dot-Product Attention (GDPA), and BlockAttention—and mixed ultra-low precision training (including MXFP8 attention and MLP) optimized for recommendation workloads. Scaling efficiency came from topology-aware 5D parallelism with Streaming Multiprocessor (SM)-free collectives, combining 2D FSDP + Expert Parallelism for dense parameters with Fully Sharded 2D Model Parallel