Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention
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_ | 1866916632592384000 |
|---|---|
| author | Yuan, Jingyang Gao, Huazuo Dai, Damai Luo, Junyu Zhao, Liang Zhang, Zhengyan Xie, Zhenda Wei, Y. X. Wang, Lean Xiao, Zhiping Wang, Yuqing Ruan, Chong Zhang, Ming Liang, Wenfeng Zeng, Wangding |
| author_facet | Yuan, Jingyang Gao, Huazuo Dai, Damai Luo, Junyu Zhao, Liang Zhang, Zhengyan Xie, Zhenda Wei, Y. X. Wang, Lean Xiao, Zhiping Wang, Yuqing Ruan, Chong Zhang, Ming Liang, Wenfeng Zeng, Wangding |
| contents | Long-context modeling is crucial for next-generation language models, yet the high computational cost of standard attention mechanisms poses significant computational challenges. Sparse attention offers a promising direction for improving efficiency while maintaining model capabilities. We present NSA, a Natively trainable Sparse Attention mechanism that integrates algorithmic innovations with hardware-aligned optimizations to achieve efficient long-context modeling. NSA employs a dynamic hierarchical sparse strategy, combining coarse-grained token compression with fine-grained token selection to preserve both global context awareness and local precision. Our approach advances sparse attention design with two key innovations: (1) We achieve substantial speedups through arithmetic intensity-balanced algorithm design, with implementation optimizations for modern hardware. (2) We enable end-to-end training, reducing pretraining computation without sacrificing model performance. As shown in Figure 1, experiments show the model pretrained with NSA maintains or exceeds Full Attention models across general benchmarks, long-context tasks, and instruction-based reasoning. Meanwhile, NSA achieves substantial speedups over Full Attention on 64k-length sequences across decoding, forward propagation, and backward propagation, validating its efficiency throughout the model lifecycle. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_11089 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention Yuan, Jingyang Gao, Huazuo Dai, Damai Luo, Junyu Zhao, Liang Zhang, Zhengyan Xie, Zhenda Wei, Y. X. Wang, Lean Xiao, Zhiping Wang, Yuqing Ruan, Chong Zhang, Ming Liang, Wenfeng Zeng, Wangding Computation and Language Artificial Intelligence Machine Learning Long-context modeling is crucial for next-generation language models, yet the high computational cost of standard attention mechanisms poses significant computational challenges. Sparse attention offers a promising direction for improving efficiency while maintaining model capabilities. We present NSA, a Natively trainable Sparse Attention mechanism that integrates algorithmic innovations with hardware-aligned optimizations to achieve efficient long-context modeling. NSA employs a dynamic hierarchical sparse strategy, combining coarse-grained token compression with fine-grained token selection to preserve both global context awareness and local precision. Our approach advances sparse attention design with two key innovations: (1) We achieve substantial speedups through arithmetic intensity-balanced algorithm design, with implementation optimizations for modern hardware. (2) We enable end-to-end training, reducing pretraining computation without sacrificing model performance. As shown in Figure 1, experiments show the model pretrained with NSA maintains or exceeds Full Attention models across general benchmarks, long-context tasks, and instruction-based reasoning. Meanwhile, NSA achieves substantial speedups over Full Attention on 64k-length sequences across decoding, forward propagation, and backward propagation, validating its efficiency throughout the model lifecycle. |
| title | Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2502.11089 |