Salvato in:
| Autori principali: | , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2506.19171 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866912446271193088 |
|---|---|
| author | Huang, Xingyue Hu, Xianglong Ding, Zifeng He, Yuan Rishabh Alzarooni, Waleed Ye, Ziyu Fan, Wendong He, Bailan Bo, Haige Hu, Changran Li, Guohao |
| author_facet | Huang, Xingyue Hu, Xianglong Ding, Zifeng He, Yuan Rishabh Alzarooni, Waleed Ye, Ziyu Fan, Wendong He, Bailan Bo, Haige Hu, Changran Li, Guohao |
| contents | Large language models (LLMs) often struggle with mathematical problems that require exact computation or multi-step algebraic reasoning. Tool-integrated reasoning (TIR) offers a promising solution by leveraging external tools such as code interpreters to ensure correctness, but it introduces inference-time dependencies that hinder scalability and deployment. In this work, we propose a new paradigm for distilling tool knowledge into LLMs purely through natural language. We first construct a Solver Agent that solves math problems by interleaving planning, symbolic tool calls, and reflective reasoning. Then, using a back-translation pipeline powered by multiple LLM-based agents, we convert interleaved TIR traces into natural language reasoning traces. A Translator Agent generates explanations for individual tool calls, while a Rephrase Agent merges them into a fluent and globally coherent narrative. Empirically, we show that fine-tuning a small open-source model on these synthesized traces enables it to internalize both tool knowledge and structured reasoning patterns, yielding gains on competition-level math benchmarks without requiring tool access at inference. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_19171 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Distilling Tool Knowledge into Language Models via Back-Translated Traces Huang, Xingyue Hu, Xianglong Ding, Zifeng He, Yuan Rishabh Alzarooni, Waleed Ye, Ziyu Fan, Wendong He, Bailan Bo, Haige Hu, Changran Li, Guohao Machine Learning Large language models (LLMs) often struggle with mathematical problems that require exact computation or multi-step algebraic reasoning. Tool-integrated reasoning (TIR) offers a promising solution by leveraging external tools such as code interpreters to ensure correctness, but it introduces inference-time dependencies that hinder scalability and deployment. In this work, we propose a new paradigm for distilling tool knowledge into LLMs purely through natural language. We first construct a Solver Agent that solves math problems by interleaving planning, symbolic tool calls, and reflective reasoning. Then, using a back-translation pipeline powered by multiple LLM-based agents, we convert interleaved TIR traces into natural language reasoning traces. A Translator Agent generates explanations for individual tool calls, while a Rephrase Agent merges them into a fluent and globally coherent narrative. Empirically, we show that fine-tuning a small open-source model on these synthesized traces enables it to internalize both tool knowledge and structured reasoning patterns, yielding gains on competition-level math benchmarks without requiring tool access at inference. |
| title | Distilling Tool Knowledge into Language Models via Back-Translated Traces |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2506.19171 |