GEM Training: How Meta Doubled the Efficiency of Its LLM-Scale Ads Foundation Model

Meta Engineering published “GEM Training: How Meta Doubled the Efficiency of Its LLM-Scale Ads Foundation Model” dated 2026-08-03.

According to the Meta Engineering feed, 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 same item also notes that this post goes into the details on how we achieved: doubling end-to-end (E2E) training efficiency to 20–25% Model FLOPs Utilization (MFU) while scaling training FLOPs 4x in [...] Read More...

The same item also notes that post GEM Training: How Meta Doubled the Efficiency of Its LLM-Scale Ads Foundation Model appeared first on Engineering at Meta.

According to the Meta Engineering feed, this post goes into the details on how we achieved: doubling end-to-end (E2E) training efficiency to 20–25% Model FLOPs Utilization (MFU) while scaling training FLOPs 4x in [...] Read More...

The same item also notes that post GEM Training: How Meta Doubled the Efficiency of Its LLM-Scale Ads Foundation Model appeared first on Engineering at Meta.

This item is filed from a public RSS feed.

Signal Desk writes its own brief and does not reprint the source article.

Open the original at https://engineering.fb.com/2026/08/03/ml-applications/training-gem-at-llm-scale-meta-ads-recommendation-foundation-model/ to read the publisher’s full post, quotes, and any figures they published.