mlsanityData to 7 Oct 2026

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

arXiv 2604.12110 ↗First posted 13 Apr 2026cs.LGcited by 2

Not on the map. Why?

Abstract (arXiv)

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.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Builds on 22 of its 39 references that are in the map. First posted on arXiv 13 Apr 2026.

Builds on

22 of 39 references.

References · 22

  1. 2015Distilling the Knowledge in a Neural Networkcited by 10,076
  2. 2019Parameter-Efficient Transfer Learning for NLPcited by 4,195
  3. 2022Fast Inference from Transformers via Speculative Decodingcited by 1,615
  4. 2021Towards a Unified View of Parameter-Efficient Transfer Learningcited by 863
  5. 2016Sequence-Level Knowledge Distillationcited by 761
  6. 2016Wide & Deep Learning for Recommender Systemscited by 745
  7. 2018Born Again Neural Networkscited by 539
  8. 2020Parameter-Efficient Transfer Learning with Diff Pruningcited by 351
  9. 2019Deep Learning Recommendation Model for Personalization and Recommendation Systemscited by 327
  10. 2020DCN V2: Improved Deep & Cross Network and Practical Lessons for Web-scale Learning to Rank Systemscited by 326
  11. 2022Towards Universal Sequence Representation Learning for Recommender Systemscited by 120
  12. 2024Wukong: Towards a Scaling Law for Large-Scale Recommendationcited by 72
10 more
  1. 2018Ranking Distillation: Learning Compact Ranking Models With High Performance for Recommender Systemcited by 60
  2. 2018Online Model Distillation for Efficient Video Inferencecited by 60
  3. 2021Semantic Models for the First-stage Retrieval: A Comprehensive Reviewcited by 57
  4. 2025Think Before Recommend: Unleashing the Latent Reasoning Power for Sequential Recommendationcited by 50
  5. 2018Learning Heterogeneous Knowledge Base Embeddings for Explainable Recommendationcited by 39
  6. 2025External Large Foundation Model: How to Efficiently Serve Trillions of Parameters for Online Ads Recommendationcited by 13
  7. 2022Sequential/Session-based Recommendations: Challenges, Approaches, Applications and Opportunitiescited by 9
  8. 2024InterFormer: Effective Heterogeneous Interaction Learning for Click-Through Rate Predictioncited by 8
  9. 2019The Pupil Has Become the Master: Teacher-Student Model-Based Word Embedding Distillation with Ensemble Learningcited by 5
  10. 2025Meta Lattice: Model Space Redesign for Cost-Effective Industry-Scale Ads Recommendationscited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026ROCS: Request-Oriented Compute Sharing for Efficient Large-Scale Recommendationcited by 0
  2. 2026SequenceO1: End-to-End Ultra-Long (100K) Sequence Modeling in Recommendation with Low-Rank Cachingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026RelayGR: Scaling Long-Sequence Generative Recommendation via Cross-Stage Relay-Race Inference
  2. 2026GEMs: Breaking the Long-Sequence Barrier in Generative Recommendation with a Multi-Stream Decoder
  3. 2025LIME: Link-based user-item Interaction Modeling with decoupled xor attention for Efficient test time scaling
  4. 2025Inference Computation Scaling for Feature Augmentation in Recommendation Systems
  5. 2025NEZHA: A Zero-sacrifice and Hyperspeed Decoding Architecture for Generative Recommendations
  6. 2026PROMISE: Process Reward Models Unlock Test-Time Scaling Laws in Generative Recommendations
  7. 2026Principled Synthetic Data Enables the First Scaling Laws for LLMs in Recommendation
  8. 2026TwiSTAR:Think Fast, Think Slow, Then Act,Generative Recommendation with Adaptive Reasoning