# SOLARIS: Speculative Offloading of Latent-bAsed Representation for Inference Scaling

Source: [SOLARIS: Speculative Offloading of Latent-bAsed Representation for Inference Scaling](<https://www.sota2.com/research/paper/69df1ba07d0172140ec7b9ba>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-06T05:00:04.315Z

Authors: Zikun Liu, Liang Luo, Qianru Li, Zhengyu Zhang, Wei Ling, Jingyi Shen, Zeliang Chen, Yaning Huang, Jingxian Huang, Abdallah Aboelela, Chonglin Sun, Feifan Gu, Fenggang Wu, Hang Qu, Huayu Li, Jill Pan, Kaidi Pei, Laming Chen, Longhao Jin, Qin Huang, Tongyi Tang, Varna Puvvada, Wenlin Chen, Xiaohan Wei, Xu Cao, Yantao Yao, Yuan Jin, Yunchen Pu, Yuxin Chen, Zijian Shen, Zhengkai Zhang, Jing Zhu, Dong Liang, Ellie Wen

Published: 2026-04-13T22:33:28.000Z

## Abstract

Recent advances in recommendation scaling laws have led to foundation models of unprecedented complexity. While these models offer superior performance, their computational demands make real-time serving impractical, often forcing practitioners to rely on knowledge distillation-compromising serving quality for efficiency. To address this challenge, we present SOLARIS (Speculative Offloading of Latent-bAsed Representation for Inference Scaling), a novel framework inspired by speculative decoding. SOLARIS proactively precomputes user-item interaction embeddings by predicting which user-item pairs are likely to appear in future requests, and asynchronously generating their foundation model representations ahead of time. This approach decouples the costly foundation model inference from the latency-critical serving path, enabling real-time knowledge transfer from models previously considered too expensive for online use. Deployed across Meta's advertising system serving billions of daily requests, SOLARIS achieves 0.67% revenue-driving top-line metrics gain, demonstrating its effectiveness at scale.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [CTR Prediction on Meta Facebook Feed (production live traffic)](<https://www.sota2.com/research/sota/ctr-prediction-on-meta-facebook-feed-production-live-traffic.md>): CTR Prediction; Meta Facebook Feed (production live traffic); Relative BCE Loss Reduction: 9.

- [CTR Prediction on Meta Facebook Reels (production live traffic)](<https://www.sota2.com/research/sota/ctr-prediction-on-meta-facebook-reels-production-live-traffic.md>): CTR Prediction; Meta Facebook Reels (production live traffic); Relative BCE Loss Reduction: 8.

- [CTR Prediction on Meta Instagram (production live traffic)](<https://www.sota2.com/research/sota/ctr-prediction-on-meta-instagram-production-live-traffic.md>): CTR Prediction; Meta Instagram (production live traffic); Relative BCE Loss Reduction: 5.

- [CTR Prediction on Meta Instagram Link Click (production live traffic)](<https://www.sota2.com/research/sota/ctr-prediction-on-meta-instagram-link-click-production-live-traffic.md>): CTR Prediction; Meta Instagram Link Click (production live traffic); Relative BCE Loss Reduction: 5.

- [CVR Prediction on Meta Facebook Feed + Reels (production live traffic)](<https://www.sota2.com/research/sota/cvr-prediction-on-meta-facebook-feed-plus-reels-production-live-traffic.md>): CVR prediction; Meta Facebook Feed + Reels (production live traffic); Relative BCE Loss Reduction (%): 10.

- [CVR Prediction on Meta Offsite Conversion (production live traffic)](<https://www.sota2.com/research/sota/cvr-prediction-on-meta-offsite-conversion-production-live-traffic.md>): CVR prediction; Meta Offsite Conversion (production live traffic); Relative BCE Loss Reduction: 5.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.12110>)

[PDF](<https://arxiv.org/pdf/2604.12110.pdf>)
