Inference-Time Budget Control for LLM Search Agents

LLM search agents increasingly rely on tools at inference time, but their trajectories are often constrained by hard limits on both tool calls and generated tokens. Under such dual budgets, better answers require not only stronger models, but also explicit control over which search action should receive the next budget unit and when the accumulated evidence is sufficient to commit a final answer. We study this problem in multi-hop question answering (QA) and formulate it as two-stage inference-time budget control. At search time, our controller assigns each feasible action a task-level Value-of-Information (VOI) score, defined as an operational estimate of marginal task value per unit budget under the current search state and remaining dual budget, and uses this score to choose among retrieval, decomposition, and answer commitment. After search, a selective evidence-grounded finalizer compares the trajectory answer with a refined candidate and rewrites only when the residual error appears to be a low-risk answer-form error. Across four multi-hop QA benchmarks, three LLM backbones, and four budget levels, the method yields positive aggregate gains over four audited baselines under the same hard dual-budget protocol. Ablations show that search-time budget control, especially budget-dependent penalty, provides the main performance gain, while answer-time control helps mainly when the retrieval path is already adequate. These results suggest that inference-time budget control for LLM search agents should govern both how budget is spent during search and how the final answer is committed.

Adaptive Top-K in SGD for Communication-Efficient Distributed Learning

Distributed stochastic gradient descent (SGD) with gradient compression has emerged as a communication-efficient solution to accelerate distributed learning. Top-K sparsification is one of the most popular gradient compression methods that sparsifies the gradient in a fixed degree during model training. However, there lacks an approach to adaptively adjust the degree of sparsification to maximize the potential of model performance or training speed. This paper addresses this issue by proposing a novel adaptive Top-K SGD framework, enabling adaptive degree of sparsification for each gradient descent step to maximize the convergence performance by exploring the trade-off between communication cost and convergence error. Firstly, we derive an upper bound of the convergence error for the adaptive sparsification scheme and the loss function. Secondly, we design the algorithm by minimizing the convergence error under the communication cost constraints. Finally, numerical results show that the proposed adaptive Top-K in SGD achieves a significantly better convergence rate compared with the state-of-the-art methods.