Z1: Efficient Test-time Scaling with Code
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866908294116802560 |
|---|---|
| author | Yu, Zhaojian Wu, Yinghao Zhao, Yilun Cohan, Arman Zhang, Xiao-Ping |
| author_facet | Yu, Zhaojian Wu, Yinghao Zhao, Yilun Cohan, Arman Zhang, Xiao-Ping |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_00810 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Z1: Efficient Test-time Scaling with Code Yu, Zhaojian Wu, Yinghao Zhao, Yilun Cohan, Arman Zhang, Xiao-Ping Computation and Language 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. |
| title | Z1: Efficient Test-time Scaling with Code |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2504.00810 |