Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2023
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2310.07075 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909216208322560 |
|---|---|
| author | Zhang, Kexun Chen, Hongqiao Li, Lei Wang, William |
| author_facet | Zhang, Kexun Chen, Hongqiao Li, Lei Wang, William |
| contents | Instruction-tuned large language models (LLMs) excel at many tasks but often fail to use external tools due to complicated and unfamiliar syntax constraints. While extensive fine-tuning and prompting can mitigate the issue, these approaches are expensive and hard to generalize. Furthermore, because syntax constraints are only learned implicitly during fine-tuning, models still make frequent syntax errors. Motivated by the fact that these constraints can be better satisfied explicitly with constrained decoding, we propose TOOLDEC, a decoding algorithm using finite state machines to force LLMs to follow tool syntax. Our experiments show that TOOLDEC eliminates all syntax errors, achieving significantly better performance on various base models and benchmarks. More surprisingly, when applied to generalist out-of-the-box LLMs such as Mistral-Instruct, TOOLDEC improves its accuracy in tool use from the initial 0% to an impressive 52%, matching the performance of specialized fine-tuned models such as ToolLLM. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2310_07075 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Don't Fine-Tune, Decode: Syntax Error-Free Tool Use via Constrained Decoding Zhang, Kexun Chen, Hongqiao Li, Lei Wang, William Computation and Language Artificial Intelligence Instruction-tuned large language models (LLMs) excel at many tasks but often fail to use external tools due to complicated and unfamiliar syntax constraints. While extensive fine-tuning and prompting can mitigate the issue, these approaches are expensive and hard to generalize. Furthermore, because syntax constraints are only learned implicitly during fine-tuning, models still make frequent syntax errors. Motivated by the fact that these constraints can be better satisfied explicitly with constrained decoding, we propose TOOLDEC, a decoding algorithm using finite state machines to force LLMs to follow tool syntax. Our experiments show that TOOLDEC eliminates all syntax errors, achieving significantly better performance on various base models and benchmarks. More surprisingly, when applied to generalist out-of-the-box LLMs such as Mistral-Instruct, TOOLDEC improves its accuracy in tool use from the initial 0% to an impressive 52%, matching the performance of specialized fine-tuned models such as ToolLLM. |
| title | Don't Fine-Tune, Decode: Syntax Error-Free Tool Use via Constrained Decoding |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2310.07075 |