test-time-computation

The paper introduces Inherit-MAS, a test-time evolution framework for LLM multi-agent systems that makes workflow and execution inheritance explicit, using a meta-model to synthesize workflows and a judge to refine them. With GPT-4o-mini and Qwen3-32B workers, it outperforms EvoAgent, EvoMAS, and TacoMAS on WorkBench and HotpotQA while cutting worker-token usage by up to 34.6%.

This paper introduces FlyBy, a selective querying framework that trains small reasoning models to diagnose knowledge bottlenecks and query stronger models, improving reasoning performance on hard problems with lower serving cost.

This paper introduces a method to control recurrent dynamics in neural networks for reliable test-time depth, analyzing dynamical regimes like settling, marginal, or drifting to improve performance on algorithmic tasks such as Sudoku and carry propagation.

A hierarchical vision-language-action model improves long-horizon robot manipulation by using world-model-guided test-time search to scale computation for high-level subtask decisions.

This article provides an in-depth explanation of the principle of reasoning effort in large language models: by adjusting the reasoning effort level (low/medium/high) on the same model, more intermediate reasoning trajectories (serial autoregressive computation) are allowed before outputting the final answer, thereby making trade-offs among answer quality, response speed, and computational cost. The key is that the model has learned to organize longer generation processes into effective computation, rather than simply increasing parameters or network layers.

Introduces Depth-Entropy Guided Sampling (DEGS), a training-free test-time method that exploits layer-wise entropy collapse in LLMs to improve reasoning without RL training, achieving competitive results with RL-posttrained models.

This paper introduces LoopCoder-v2, a family of 7B parameter parallel loop transformers for code generation, and studies the optimal number of loops, finding that two loops yield significant gains while more loops cause degradation.

This paper introduces Test-Time Personalization (TTP), a framework that improves LLM personalization by scaling inference-time computation through candidate sampling and reward-based selection. It diagnoses failure modes in standard reward models and proposes a probabilistic personalized reward model to mitigate them.

This paper introduces PaT (Planning-after-Trial), an adaptive test-time computation strategy for code generation that reduces inference costs by approximately 69% while maintaining performance comparable to larger models.

This paper introduces 'prefix consistency,' a method that weights candidate responses in Chain-of-Thought reasoning based on answer reproduction rates during trace regeneration. It achieves high accuracy with significantly fewer tokens than standard majority voting across various reasoning models and benchmarks.