Salvato in:
| Autori principali: | , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2503.07826 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866913728765624320 |
|---|---|
| author | Yin, Fan Wang, Zifeng Hsu, I-Hung Yan, Jun Jiang, Ke Chen, Yanfei Gu, Jindong Le, Long T. Chang, Kai-Wei Lee, Chen-Yu Palangi, Hamid Pfister, Tomas |
| author_facet | Yin, Fan Wang, Zifeng Hsu, I-Hung Yan, Jun Jiang, Ke Chen, Yanfei Gu, Jindong Le, Long T. Chang, Kai-Wei Lee, Chen-Yu Palangi, Hamid Pfister, Tomas |
| contents | Large language models (LLMs) have exhibited the ability to effectively utilize external tools to address user queries. However, their performance may be limited in complex, multi-turn interactions involving users and multiple tools. To address this, we propose Magnet, a principled framework for synthesizing high-quality training trajectories to enhance the function calling capability of large language model agents in multi-turn conversations with humans. The framework is based on automatic and iterative translations from a function signature path to a sequence of queries and executable function calls. We model the complicated function interactions in multi-turn cases with graph and design novel node operations to build reliable signature paths. Motivated by context distillation, when guiding the generation of positive and negative trajectories using a teacher model, we provide reference function call sequences as positive hints in context and contrastive, incorrect function calls as negative hints. Experiments show that training with the positive trajectories with supervised fine-tuning and preference optimization against negative trajectories, our 14B model, Magnet-14B-mDPO, obtains 68.01 on BFCL-v3 and 73.30 on ToolQuery, surpassing the performance of the teacher model Gemini-1.5-pro-002 by a large margin in function calling. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_07826 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Magnet: Multi-turn Tool-use Data Synthesis and Distillation via Graph Translation Yin, Fan Wang, Zifeng Hsu, I-Hung Yan, Jun Jiang, Ke Chen, Yanfei Gu, Jindong Le, Long T. Chang, Kai-Wei Lee, Chen-Yu Palangi, Hamid Pfister, Tomas Computation and Language Large language models (LLMs) have exhibited the ability to effectively utilize external tools to address user queries. However, their performance may be limited in complex, multi-turn interactions involving users and multiple tools. To address this, we propose Magnet, a principled framework for synthesizing high-quality training trajectories to enhance the function calling capability of large language model agents in multi-turn conversations with humans. The framework is based on automatic and iterative translations from a function signature path to a sequence of queries and executable function calls. We model the complicated function interactions in multi-turn cases with graph and design novel node operations to build reliable signature paths. Motivated by context distillation, when guiding the generation of positive and negative trajectories using a teacher model, we provide reference function call sequences as positive hints in context and contrastive, incorrect function calls as negative hints. Experiments show that training with the positive trajectories with supervised fine-tuning and preference optimization against negative trajectories, our 14B model, Magnet-14B-mDPO, obtains 68.01 on BFCL-v3 and 73.30 on ToolQuery, surpassing the performance of the teacher model Gemini-1.5-pro-002 by a large margin in function calling. |
| title | Magnet: Multi-turn Tool-use Data Synthesis and Distillation via Graph Translation |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2503.07826 |