Z1: Efficient Test-time Scaling with Code

Zhaojian Yu, Yinghao Wu, Yilun Zhao, Arman Cohan, Xiao-Ping Zhang

Z1: Efficient Test-time Scaling with Code: 27 upvotes on Hugging Face Daily Papers, #9 of 29 papers on 2025-04-02. Day-by-day upvote history.

Large Language Models (LLMs) can achieve enhanced complex problem-solving through test-time computing scaling, yet this often entails longer contexts and numerous reasoning token costs. In this paper, we propose an efficient test-time scaling method that trains LLMs on code-related reasoning trajectories, facilitating their reduction of excess thinking tokens while maintaining performance. First, we create Z1-Code-Reasoning-107K, a curated dataset of simple and complex coding problems paired with their short and long solution trajectories. Second, we present a novel Shifted Thinking Window to mitigate overthinking overhead by removing context-delimiting tags (e.g., <think>. . . </think>) and capping reasoning tokens. Trained with long and short trajectory data and equipped with Shifted Thinking Window, our model, Z1-7B, demonstrates the ability to adjust its reasoning level as the complexity of problems and exhibits efficient test-time scaling across different reasoning tasks that matches R1-Distill-Qwen-7B performance with about 30% of its average thinking tokens. Notably, fine-tuned with only code trajectories, Z1-7B demonstrates generalization to broader reasoning tasks (47.5% on GPQA Diamond). Our analysis of efficient reasoning elicitation also provides valuable insights for future research.

Paper page on Hugging Face · arXiv

Data: hysts-bot-data/daily-papers-stats and the Daily Papers API. Open data: tardellirs/paper-pulse-data. Sister project: Model Pulse, the download history of every model on the Hub.