The Workload-Router-Pool Architecture for LLM Inference Optimization: A Vision Paper from the vLLM Semantic Router Project

Over the past year, the vLLM Semantic Router project has released a series of work spanning: (1) core routing mechanisms -- signal-driven routing, context-length pool routing, router performance engineering, policy conflict detection, low-latency embedding models, category-aware semantic caching, user-feedback-driven routing adaptation, hallucination detection, and hierarchical content-safety classification for privacy and jailbreak protection; (2) fleet optimization -- fleet provisioning and energy-efficiency analysis; (3) agentic and multimodal routing -- multimodal agent routing, tool selection, CUA security, and multi-turn context memory and safety; (4) governance and standards -- inference routing protocols and multi-provider API extensions. Each paper tackled a specific problem in LLM inference, but the problems are not independent; for example, fleet provisioning depends on the routing policy, which depends on the workload mix, shifting as organizations adopt agentic and multimodal workloads. This paper distills those results into the Workload-Router-Pool (WRP) architecture, a three-dimensional framework for LLM inference optimization. Workload characterizes what the fleet serves (chat vs. agent, single-turn vs. multi-turn, warm vs. cold, prefill-heavy vs. decode-heavy). Router determines how each request is dispatched (static semantic rules, online bandit adaptation, RL-based model selection, quality-aware cascading). Pool defines where inference runs (homogeneous vs. heterogeneous GPU, disaggregated prefill/decode, KV-cache topology). We map our prior work onto a 3x3 WRP interaction matrix, identify which cells we have covered and which remain open, and propose twenty-one concrete research directions at the intersections, each grounded in our prior measurements, tiered by maturity from engineering-ready to open research.

RubricBench: Aligning Model-Generated Rubrics with Human Standards

As Large Language Model (LLM) alignment evolves from simple completions to complex, highly sophisticated generation, Reward Models are increasingly shifting toward rubric-guided evaluation to mitigate surface-level biases. However, the community lacks a unified benchmark to assess this evaluation paradigm, as existing benchmarks lack both the discriminative complexity and the ground-truth rubric annotations required for rigorous analysis. To bridge this gap, we introduce RubricBench, a curated benchmark with 1,147 pairwise comparisons specifically designed to assess the reliability of rubric-based evaluation. Our construction employs a multi-dimensional filtration pipeline to target hard samples featuring nuanced input complexity and misleading surface bias, augmenting each with expert-annotated, atomic rubrics derived strictly from instructions. Comprehensive experiments reveal a substantial capability gap between human-annotated and model-generated rubrics, indicating that even state-of-the-art models struggle to autonomously specify valid evaluation criteria, lagging considerably behind human-guided performance.

Give Users the Wheel: Towards Promptable Recommendation Paradigm

Conventional sequential recommendation models have achieved remarkable success in mining implicit behavioral patterns. However, these architectures remain structurally blind to explicit user intent: they struggle to adapt when a user's immediate goal (e.g., expressed via a natural language prompt) deviates from their historical habits. While Large Language Models (LLMs) offer the semantic reasoning to interpret such intent, existing integration paradigms force a dilemma: LLM-as-a-recommender paradigm sacrifices the efficiency and collaborative precision of ID-based retrieval, while Reranking methods are inherently bottlenecked by the recall capabilities of the underlying model. In this paper, we propose Decoupled Promptable Sequential Recommendation (DPR), a model-agnostic framework that empowers conventional sequential backbones to natively support Promptable Recommendation, the ability to dynamically steer the retrieval process using natural language without abandoning collaborative signals. DPR modulates the latent user representation directly within the retrieval space. To achieve this, we introduce a Fusion module to align the collaborative and semantic signals, a Mixture-of-Experts (MoE) architecture that disentangles the conflicting gradients from positive and negative steering, and a three-stage training strategy that progressively aligns the semantic space of prompts with the collaborative space. Extensive experiments on real-world datasets demonstrate that DPR significantly outperforms state-of-the-art baselines in prompt-guided tasks while maintaining competitive performance in standard sequential recommendation scenarios.

Optimizing User Profiles via Contextual Bandits for Retrieval-Augmented LLM Personalization

Large Language Models (LLMs) excel at general-purpose tasks, yet adapting their responses to individual users remains challenging. Retrieval augmentation provides a lightweight alternative to fine-tuning by conditioning LLMs on user history records, and existing approaches typically select these records based on semantic relevance. We argue that relevance serves as an unreliable proxy for utility: a record may be semantically similar to a query yet fail to improve generation quality or even degrade it due to redundancy or conflicting information. To bridge this gap, we propose PURPLE, a contextual bandit framework that oPtimizes UseR Profiles for Llm pErsonalization. In contrast to a greedy selection of the most relevant records, PURPLE treats profile construction as a set generation process and utilizes a Plackett-Luce ranking model to capture complex inter-record dependencies. By training with dense feedback provided by the likelihood of the reference response, our method aligns retrieval directly with generation quality. Extensive experiments on nine personalization tasks demonstrate that PURPLE consistently outperforms strong heuristic and retrieval-augmented baselines in both effectiveness and efficiency, establishing a principled and scalable solution for optimizing user profiles.

Exploring Test-time Scaling via Prediction Merging on Large-Scale Recommendation

Inspired by the success of language models (LM), scaling up deep learning recommendation systems (DLRS) has become a recent trend in the community. All previous methods tend to scale up the model parameters during training time. However, how to efficiently utilize and scale up computational resources during test time remains underexplored, which can prove to be a scaling-efficient approach and bring orthogonal improvements in LM domains. The key point in applying test-time scaling to DLRS lies in effectively generating diverse yet meaningful outputs for the same instance. We propose two ways: One is to explore the heterogeneity of different model architectures. The other is to utilize the randomness of model initialization under a homogeneous architecture. The evaluation is conducted across eight models, including both classic and SOTA models, on three benchmarks. Sufficient evidence proves the effectiveness of both solutions. We further prove that under the same inference budget, test-time scaling can outperform parameter scaling. Our test-time scaling can also be seamlessly accelerated with the increase in parallel servers when deployed online, without affecting the inference time on the user side. Code is available.

Timing is important: Risk-aware Fund Allocation based on Time-Series Forecasting

Fund allocation has been an increasingly important problem in the financial domain. In reality, we aim to allocate the funds to buy certain assets within a certain future period. Naive solutions such as prediction-only or Predict-then-Optimize approaches suffer from goal mismatch. Additionally, the introduction of the SOTA time series forecasting model inevitably introduces additional uncertainty in the predicted result. To solve both problems mentioned above, we introduce a Risk-aware Time-Series Predict-and-Allocate (RTS-PnO) framework, which holds no prior assumption on the forecasting models. Such a framework contains three features: (i) end-to-end training with objective alignment measurement, (ii) adaptive forecasting uncertainty calibration, and (iii) agnostic towards forecasting models. The evaluation of RTS-PnO is conducted over both online and offline experiments. For offline experiments, eight datasets from three categories of financial applications are used: Currency, Stock, and Cryptos. RTS-PnO consistently outperforms other competitive baselines. The online experiment is conducted on the Cross-Border Payment business at FiT, Tencent, and an 8.4\% decrease in regret is witnessed when compared with the product-line approach. The code for the offline experiment is available at https://github.com/fuyuanlyu/RTS-PnO.

What, How, Where, and How Well? A Survey on Test-Time Scaling in Large Language Models

As enthusiasm for scaling computation (data and parameters) in the pretraining era gradually diminished, test-time scaling (TTS), also referred to as ``test-time computing'' has emerged as a prominent research focus. Recent studies demonstrate that TTS can further elicit the problem-solving capabilities of large language models (LLMs), enabling significant breakthroughs not only in specialized reasoning tasks, such as mathematics and coding, but also in general tasks like open-ended Q&A. However, despite the explosion of recent efforts in this area, there remains an urgent need for a comprehensive survey offering a systemic understanding. To fill this gap, we propose a unified, multidimensional framework structured along four core dimensions of TTS research: what to scale, how to scale, where to scale, and how well to scale. Building upon this taxonomy, we conduct an extensive review of methods, application scenarios, and assessment aspects, and present an organized decomposition that highlights the unique functional roles of individual techniques within the broader TTS landscape. From this analysis, we distill the major developmental trajectories of TTS to date and offer hands-on guidelines for practical deployment. Furthermore, we identify several open challenges and offer insights into promising future directions, including further scaling, clarifying the functional essence of techniques, generalizing to more tasks, and more attributions.

A Predict-Then-Optimize Customer Allocation Framework for Online Fund Recommendation

With the rapid growth of online investment platforms, funds can be distributed to individual customers online. The central issue is to match funds with potential customers under constraints. Most mainstream platforms adopt the recommendation formulation to tackle the problem. However, the traditional recommendation regime has its inherent drawbacks when applying the fund-matching problem with multiple constraints. In this paper, we model the fund matching under the allocation formulation. We design PTOFA, a Predict-Then-Optimize Fund Allocation framework. This data-driven framework consists of two stages, i.e., prediction and optimization, which aim to predict expected revenue based on customer behavior and optimize the impression allocation to achieve the maximum revenue under the necessary constraints, respectively. Extensive experiments on real-world datasets from an industrial online investment platform validate the effectiveness and efficiency of our solution. Additionally, the online A/B tests demonstrate PTOFA's effectiveness in the real-world fund recommendation scenario.

Crowd Comparative Reasoning: Unlocking Comprehensive Evaluations for LLM-as-a-Judge

LLM-as-a-Judge, which generates chain-of-thought (CoT) judgments, has become a widely adopted auto-evaluation method. However, its reliability is compromised by the CoT reasoning's inability to capture comprehensive and deeper details, often leading to incomplete outcomes. Existing methods mainly rely on majority voting or criteria expansion, which is insufficient to address the limitation in CoT. We propose Crowd-based Comparative Evaluation, which introduces additional crowd responses to compare with the candidate responses, thereby exposing deeper and more comprehensive details within the candidate responses. This process effectively guides LLM-as-a-Judge to provide a more detailed CoT judgment. Extensive experiments demonstrate that our approach enhances evaluation reliability, achieving an average accuracy gain of 6.7% across five benchmarks. Moreover, our method produces higher-quality CoTs that facilitate judge distillation and exhibit superior performance in rejection sampling for supervised fine-tuning (SFT), referred to as crowd rejection sampling, thereby enabling more efficient SFT. Our analysis confirms that CoTs generated by ours are more comprehensive and of higher quality, and evaluation accuracy improves as inference scales.

Fusion Matters: Learning Fusion in Deep Click-through Rate Prediction Models

The evolution of previous Click-Through Rate (CTR) models has mainly been driven by proposing complex components, whether shallow or deep, that are adept at modeling feature interactions. However, there has been less focus on improving fusion design. Instead, two naive solutions, stacked and parallel fusion, are commonly used. Both solutions rely on pre-determined fusion connections and fixed fusion operations. It has been repetitively observed that changes in fusion design may result in different performances, highlighting the critical role that fusion plays in CTR models. While there have been attempts to refine these basic fusion strategies, these efforts have often been constrained to specific settings or dependent on specific components. Neural architecture search has also been introduced to partially deal with fusion design, but it comes with limitations. The complexity of the search space can lead to inefficient and ineffective results. To bridge this gap, we introduce OptFusion, a method that automates the learning of fusion, encompassing both the connection learning and the operation selection. We have proposed a one-shot learning algorithm tackling these tasks concurrently. Our experiments are conducted over three large-scale datasets. Extensive experiments prove both the effectiveness and efficiency of OptFusion in improving CTR model performance. Our code implementation is available here\url{https://github.com/kexin-kxzhang/OptFusion}.