TARo: Token-level Adaptive Routing for LLM Test-time Alignment

• URL: http://arxiv.org/abs/2603.18411v1
• Date: Thu, 19 Mar 2026 02:18:40 GMT
• Title: TARo: Token-level Adaptive Routing for LLM Test-time Alignment
• Abstract summary: TARo steers frozen large language models toward structured reasoning entirely at inference time.<n>Extensive experiments show that TARo significantly improves reasoning performance by up to +22.4% over base model.<n> TARo also generalizes from small to large backbones without retraining.
• Score: 15.824259013420807
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large language models (LLMs) exhibit strong reasoning capabilities but typically require expensive post-training to reach high performance. Recent test-time alignment methods offer a lightweight alternative, but have been explored mainly for preference alignment rather than reasoning. To bridge this gap, we propose, Token-level Adaptive Routing (TARo), which steers frozen LLMs toward structured reasoning entirely at inference time. Specifically, we first train reward models on step-wise mathematical traces to capture fine-grained logical consistency signals, then introduce a learnable token-level router that automatically controls the guidance of the reward model to the base model. Extensive experiments show that TARo significantly improves reasoning performance by up to +22.4% over base model and +8.4% over existing token-level test-time alignment methods, while also boosting out-of-distribution clinical reasoning (MedXpertQA) and instruction following (AlpacaEval). Furthermore, TARo also generalizes from small to large backbones without retraining, extending test-time alignment from preference optimization to robust, cross-domain reasoning.

Related papers

• Stop Thinking, Start Looking: Efficient Post-Training for Multimodal Document Question Answering via Reasoning-Free Alignment [2.8730465903425877]
  We introduce Perception-RFT, a training framework that applies Group Relative Policy Optimization to multimodal document QA.<n>We find that reasoning-enabled models suppress their reasoning traces during training, converging to direct perception-based policies at the 4B parameter scale.<n>We show that an early SFT$rightarrow$RL transition achieves comparable precision with 65% less training data.
  arXiv  Detail & Related papers  (2026-07-16T07:43:39Z)
• Intuition-Guided Latent Reasoning for LLM-Based Recommendation [72.06271644558291]
  We propose emphIntuRec, a framework that anchors latent reasoning with emphrecommendation intuition.<n>IntuRec consistently outperforms state-of-the-art baselines in experiments on multiple real-world datasets.
  arXiv  Detail & Related papers  (2026-06-26T03:29:48Z)
• REVES: REvision and VErification--Augmented Training for Test-Time Scaling [53.197756110943395]
  We propose a two-stage iterative framework that alternates between online data/prompt augmentation and policy optimization.<n>We observe gains of +6.5 points over the RL baseline and +4.0 points over standard multi-turn training.
  arXiv  Detail & Related papers  (2026-06-17T10:37:23Z)
• $\nabla$-Reasoner: LLM Reasoning via Test-Time Gradient Descent in Latent Space [71.23672814629448]
  $nabla$-Reasoner is an iterative generation framework that integrates differentiable optimization over token logits into the decoding loop.<n>$nabla$-Reasoner achieves over 20% accuracy improvement on a challenging mathematical reasoning benchmark.
  arXiv  Detail & Related papers  (2026-03-05T08:42:54Z)
• Reasoning Pattern Alignment Merging for Adaptive Reasoning [48.347817456299104]
  Reasoning Pattern Alignment Merging (RPAM)<n>RPAM is a layer-wise model merging framework based on feature alignment to facilitate query-adaptive reasoning.<n> Experiments on seven widely used reasoning benchmarks show that RPAM substantially reduces inference cost while maintaining strong performance.
  arXiv  Detail & Related papers  (2026-01-07T01:36:39Z)
• Learning to Ponder: Adaptive Reasoning in Latent Space [2.8835557003761747]
  We present FR-Ponder, a single-graph, backbone-training-free framework that allocates instance-adaptive reasoning compute via latent steering.<n>On GSM8K and MATH500, FR-Ponder improves the compute-accuracy frontier, delivering lower FLOPs with better matched accuracy and comparing favorably to early-exit baselines.
  arXiv  Detail & Related papers  (2025-09-29T03:21:42Z)
• PITA: Preference-Guided Inference-Time Alignment for LLM Post-Training [9.093854840532062]
  PITA is a novel framework that integrates preference feedback directly into the LLM's token generation.<n> PITA learns a small preference-based guidance policy to modify token probabilities at inference time without fine-tuning.<n>We evaluate PITA across diverse tasks, including mathematical reasoning and sentiment classification.
  arXiv  Detail & Related papers  (2025-07-26T21:46:32Z)
• Aligning Frozen LLMs by Reinforcement Learning: An Iterative Reweight-then-Optimize Approach [65.6966065843227]
  Iterative Reweight-then-IRO is a framework that performs RL-style alignment of a frozen base model without touching its parameters.<n>At test time, the value functions are used to guide the base model generation via a search-based optimization process.<n> Notably, users can apply IRO to align a model on their own dataset, similar to OpenAI's reinforcement fine-tuning (RFT)
  arXiv  Detail & Related papers  (2025-06-21T21:49:02Z)
• Token-Efficient RL for LLM Reasoning [0.02488650627593658]
  We propose reinforcement learning strategies tailored for reasoning in large language models (LLMs) under strict memory and compute limits.<n>Building on early policy gradient methods with baseline subtraction, we design critic-free methods that operate on a small, informative subset of output tokens.<n>We show that our methods raise accuracy on the SVAMP benchmark from 46% to over 70% and show strong performance on multi-digit multiplication.
  arXiv  Detail & Related papers  (2025-04-29T14:58:43Z)
• The First Few Tokens Are All You Need: An Efficient and Effective Unsupervised Prefix Fine-Tuning Method for Reasoning Models [69.798277882245]
  We introduce Unsupervised Prefix Fine-Tuning (UPFT) to enhance large language models' reasoning efficiency.<n>UPFT removes the need for labeled data or exhaustive sampling.<n> Experiments show that UPFT matches the performance of supervised methods.
  arXiv  Detail & Related papers  (2025-03-04T18:56:03Z)
• S$^2$R: Teaching LLMs to Self-verify and Self-correct via Reinforcement Learning [51.84977135926156]
  We introduce S$2$R, an efficient framework that enhances LLM reasoning by teaching models to self-verify and self-correct during inference.<n>Our results demonstrate that Qwen2.5-math-7B achieves an accuracy improvement from 51.0% to 81.6%, outperforming models trained on an equivalent amount of long-CoT distilled data.
  arXiv  Detail & Related papers  (2025-02-18T13:40:22Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.