Answer, Assemble, Ace: Understanding How LMs Answer Multiple Choice Questions
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915186885001216 |
|---|---|
| author | Wiegreffe, Sarah Tafjord, Oyvind Belinkov, Yonatan Hajishirzi, Hannaneh Sabharwal, Ashish |
| author_facet | Wiegreffe, Sarah Tafjord, Oyvind Belinkov, Yonatan Hajishirzi, Hannaneh Sabharwal, Ashish |
| contents | Multiple-choice question answering (MCQA) is a key competence of performant transformer language models that is tested by mainstream benchmarks. However, recent evidence shows that models can have quite a range of performance, particularly when the task format is diversified slightly (such as by shuffling answer choice order). In this work we ask: how do successful models perform formatted MCQA? We employ vocabulary projection and activation patching methods to localize key hidden states that encode relevant information for predicting the correct answer. We find that the prediction of a specific answer symbol is causally attributed to a few middle layers, and specifically their multi-head self-attention mechanisms. We show that subsequent layers increase the probability of the predicted answer symbol in vocabulary space, and that this probability increase is associated with a sparse set of attention heads with unique roles. We additionally uncover differences in how different models adjust to alternative symbols. Finally, we demonstrate that a synthetic task can disentangle sources of model error to pinpoint when a model has learned formatted MCQA, and show that logit differences between answer choice tokens continue to grow over the course of training. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2407_15018 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Answer, Assemble, Ace: Understanding How LMs Answer Multiple Choice Questions Wiegreffe, Sarah Tafjord, Oyvind Belinkov, Yonatan Hajishirzi, Hannaneh Sabharwal, Ashish Computation and Language Artificial Intelligence Multiple-choice question answering (MCQA) is a key competence of performant transformer language models that is tested by mainstream benchmarks. However, recent evidence shows that models can have quite a range of performance, particularly when the task format is diversified slightly (such as by shuffling answer choice order). In this work we ask: how do successful models perform formatted MCQA? We employ vocabulary projection and activation patching methods to localize key hidden states that encode relevant information for predicting the correct answer. We find that the prediction of a specific answer symbol is causally attributed to a few middle layers, and specifically their multi-head self-attention mechanisms. We show that subsequent layers increase the probability of the predicted answer symbol in vocabulary space, and that this probability increase is associated with a sparse set of attention heads with unique roles. We additionally uncover differences in how different models adjust to alternative symbols. Finally, we demonstrate that a synthetic task can disentangle sources of model error to pinpoint when a model has learned formatted MCQA, and show that logit differences between answer choice tokens continue to grow over the course of training. |
| title | Answer, Assemble, Ace: Understanding How LMs Answer Multiple Choice Questions |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2407.15018 |