Attention Sinks Induce Gradient Sinks: Massive Activations as Gradient Regulators in Transformers
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914533166022656 |
|---|---|
| author | Chen, Yihong Lin, Zhouchen Yao, Quanming |
| author_facet | Chen, Yihong Lin, Zhouchen Yao, Quanming |
| contents | Attention sinks and massive activations are recurring and closely related phenomena in Transformer models. Existing explanations have largely focused on the forward pass, yet in pre-norm Transformers, large residual-stream norms play only an indirect forward role because sublayers operate on normalized inputs. We study this relationship from the perspective of backpropagation. Empirically and theoretically, we show that under causal masking, attention sinks can induce pronounced gradient concentration, which we term gradient sinks. Since the RMSNorm Jacobian attenuates gradients roughly in inverse proportion to input norm, massive activations can be understood as adaptive regulators of this localized gradient pressure during training. This interpretation predicts that attenuating sink-induced gradients should weaken massive activations. We test this prediction with V-scale, a modification that adjusts backpropagated gradients on the value path. In V-scale models, attention sinks are preserved, whereas massive activations are suppressed. These results identify gradient sinks as a backward-pass counterpart of attention sinks, and massive activations as an adaptive RMSNorm-mediated response that attenuates the resulting localized training pressure. Our code is available at https://anonymous.4open.science/r/GradientSinkCode-B309. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_17771 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Attention Sinks Induce Gradient Sinks: Massive Activations as Gradient Regulators in Transformers Chen, Yihong Lin, Zhouchen Yao, Quanming Machine Learning Artificial Intelligence Attention sinks and massive activations are recurring and closely related phenomena in Transformer models. Existing explanations have largely focused on the forward pass, yet in pre-norm Transformers, large residual-stream norms play only an indirect forward role because sublayers operate on normalized inputs. We study this relationship from the perspective of backpropagation. Empirically and theoretically, we show that under causal masking, attention sinks can induce pronounced gradient concentration, which we term gradient sinks. Since the RMSNorm Jacobian attenuates gradients roughly in inverse proportion to input norm, massive activations can be understood as adaptive regulators of this localized gradient pressure during training. This interpretation predicts that attenuating sink-induced gradients should weaken massive activations. We test this prediction with V-scale, a modification that adjusts backpropagated gradients on the value path. In V-scale models, attention sinks are preserved, whereas massive activations are suppressed. These results identify gradient sinks as a backward-pass counterpart of attention sinks, and massive activations as an adaptive RMSNorm-mediated response that attenuates the resulting localized training pressure. Our code is available at https://anonymous.4open.science/r/GradientSinkCode-B309. |
| title | Attention Sinks Induce Gradient Sinks: Massive Activations as Gradient Regulators in Transformers |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2603.17771 |