Quantifying Logical Consistency in Transformers via Query-Key Alignment
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866917956526538752 |
|---|---|
| author | Tulchinskii, Eduard Voznyuk, Anastasia Kushnareva, Laida Andriiainen, Andrei Piontkovskaya, Irina Burnaev, Evgeny Barannikov, Serguei |
| author_facet | Tulchinskii, Eduard Voznyuk, Anastasia Kushnareva, Laida Andriiainen, Andrei Piontkovskaya, Irina Burnaev, Evgeny Barannikov, Serguei |
| contents | Large language models (LLMs) have demonstrated impressive performance in various natural language processing tasks, yet their ability to perform multi-step logical reasoning remains an open challenge. Although Chain-of-Thought prompting has improved logical reasoning by enabling models to generate intermediate steps, it lacks mechanisms to assess the coherence of these logical transitions. In this paper, we propose a novel, lightweight evaluation strategy for logical reasoning that uses query-key alignments inside transformer attention heads. By computing a single forward pass and extracting a "QK-score" from carefully chosen heads, our method reveals latent representations that reliably separate valid from invalid inferences, offering a scalable alternative to traditional ablation-based techniques. We also provide an empirical validation on multiple logical reasoning benchmarks, demonstrating improved robustness of our evaluation method against distractors and increased reasoning depth. The experiments were conducted on a diverse set of models, ranging from 1.5B to 70B parameters. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_17017 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Quantifying Logical Consistency in Transformers via Query-Key Alignment Tulchinskii, Eduard Voznyuk, Anastasia Kushnareva, Laida Andriiainen, Andrei Piontkovskaya, Irina Burnaev, Evgeny Barannikov, Serguei Computation and Language Information Theory Machine Learning Logic Large language models (LLMs) have demonstrated impressive performance in various natural language processing tasks, yet their ability to perform multi-step logical reasoning remains an open challenge. Although Chain-of-Thought prompting has improved logical reasoning by enabling models to generate intermediate steps, it lacks mechanisms to assess the coherence of these logical transitions. In this paper, we propose a novel, lightweight evaluation strategy for logical reasoning that uses query-key alignments inside transformer attention heads. By computing a single forward pass and extracting a "QK-score" from carefully chosen heads, our method reveals latent representations that reliably separate valid from invalid inferences, offering a scalable alternative to traditional ablation-based techniques. We also provide an empirical validation on multiple logical reasoning benchmarks, demonstrating improved robustness of our evaluation method against distractors and increased reasoning depth. The experiments were conducted on a diverse set of models, ranging from 1.5B to 70B parameters. |
| title | Quantifying Logical Consistency in Transformers via Query-Key Alignment |
| topic | Computation and Language Information Theory Machine Learning Logic |
| url | https://arxiv.org/abs/2502.17017 |