Symbol tuning improves in-context learning in language models
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2023
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910284072878080 |
|---|---|
| author | Wei, Jerry Hou, Le Lampinen, Andrew Chen, Xiangning Huang, Da Tay, Yi Chen, Xinyun Lu, Yifeng Zhou, Denny Ma, Tengyu Le, Quoc V. |
| author_facet | Wei, Jerry Hou, Le Lampinen, Andrew Chen, Xiangning Huang, Da Tay, Yi Chen, Xinyun Lu, Yifeng Zhou, Denny Ma, Tengyu Le, Quoc V. |
| contents | We present symbol tuning - finetuning language models on in-context input-label pairs where natural language labels (e.g., "positive/negative sentiment") are replaced with arbitrary symbols (e.g., "foo/bar"). Symbol tuning leverages the intuition that when a model cannot use instructions or natural language labels to figure out a task, it must instead do so by learning the input-label mappings.
We experiment with symbol tuning across Flan-PaLM models up to 540B parameters and observe benefits across various settings. First, symbol tuning boosts performance on unseen in-context learning tasks and is much more robust to underspecified prompts, such as those without instructions or without natural language labels. Second, symbol-tuned models are much stronger at algorithmic reasoning tasks, with up to 18.2% better performance on the List Functions benchmark and up to 15.3% better performance on the Simple Turing Concepts benchmark. Finally, symbol-tuned models show large improvements in following flipped-labels presented in-context, meaning that they are more capable of using in-context information to override prior semantic knowledge. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2305_08298 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Symbol tuning improves in-context learning in language models Wei, Jerry Hou, Le Lampinen, Andrew Chen, Xiangning Huang, Da Tay, Yi Chen, Xinyun Lu, Yifeng Zhou, Denny Ma, Tengyu Le, Quoc V. Computation and Language We present symbol tuning - finetuning language models on in-context input-label pairs where natural language labels (e.g., "positive/negative sentiment") are replaced with arbitrary symbols (e.g., "foo/bar"). Symbol tuning leverages the intuition that when a model cannot use instructions or natural language labels to figure out a task, it must instead do so by learning the input-label mappings. We experiment with symbol tuning across Flan-PaLM models up to 540B parameters and observe benefits across various settings. First, symbol tuning boosts performance on unseen in-context learning tasks and is much more robust to underspecified prompts, such as those without instructions or without natural language labels. Second, symbol-tuned models are much stronger at algorithmic reasoning tasks, with up to 18.2% better performance on the List Functions benchmark and up to 15.3% better performance on the Simple Turing Concepts benchmark. Finally, symbol-tuned models show large improvements in following flipped-labels presented in-context, meaning that they are more capable of using in-context information to override prior semantic knowledge. |
| title | Symbol tuning improves in-context learning in language models |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2305.08298 |