Short-Range Dependency Effects on Transformer Instability and a Decomposed Attention Solution
Fuente:
arXiv
Enregistré dans:
| Auteur principal: | |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866916749506510848 |
|---|---|
| author | Hajra, Suvadeep |
| author_facet | Hajra, Suvadeep |
| contents | Transformer language models have driven significant progress across various fields, including natural language processing and computer vision. A central component of these models is the self-attention (SA) mechanism, which learns rich vector representations of tokens by modeling their relationships with others in a sequence. However, despite extensive research, transformers continue to suffer from training instability -- often manifesting as spikes or divergence in the training loss during a run.
In this work, we identify one source of this instability: SA's limited ability to capture short-range dependencies, especially in tasks like language modeling, where almost every token heavily relies on its nearby neighbors. This limitation causes the pre-softmax logits of SA to grow rapidly, destabilizing training. To address this, we propose decomposing the SA into local (short-range) and global (long-range) attention heads. This decomposed attention, referred to as Long Short-attention (LS-attention), mitigates logit explosion and results in more stable training compared to an equivalent multi-head self-attention (MHSA). Empirical comparisons with two alternative training stabilization methods show that LS-attention reduces the validation perplexity to nearly 2/5 of that achieved by one method and reaches a similar perplexity as the other method using only 1/20 of the GPU hours. Additionally, our experiments demonstrate that LS-attention reduces inference latency by up to 36% compared to a state-of-the-art implementation of equivalent MHSA. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_15548 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Short-Range Dependency Effects on Transformer Instability and a Decomposed Attention Solution Hajra, Suvadeep Machine Learning Transformer language models have driven significant progress across various fields, including natural language processing and computer vision. A central component of these models is the self-attention (SA) mechanism, which learns rich vector representations of tokens by modeling their relationships with others in a sequence. However, despite extensive research, transformers continue to suffer from training instability -- often manifesting as spikes or divergence in the training loss during a run. In this work, we identify one source of this instability: SA's limited ability to capture short-range dependencies, especially in tasks like language modeling, where almost every token heavily relies on its nearby neighbors. This limitation causes the pre-softmax logits of SA to grow rapidly, destabilizing training. To address this, we propose decomposing the SA into local (short-range) and global (long-range) attention heads. This decomposed attention, referred to as Long Short-attention (LS-attention), mitigates logit explosion and results in more stable training compared to an equivalent multi-head self-attention (MHSA). Empirical comparisons with two alternative training stabilization methods show that LS-attention reduces the validation perplexity to nearly 2/5 of that achieved by one method and reaches a similar perplexity as the other method using only 1/20 of the GPU hours. Additionally, our experiments demonstrate that LS-attention reduces inference latency by up to 36% compared to a state-of-the-art implementation of equivalent MHSA. |
| title | Short-Range Dependency Effects on Transformer Instability and a Decomposed Attention Solution |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2505.15548 |