A Silver Bullet or a Compromise for Full Attention? A Comprehensive Study of Gist Token-based Context Compression
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866912167363608576 |
|---|---|
| author | Deng, Chenlong Zhang, Zhisong Mao, Kelong Li, Shuaiyi Huang, Xinting Yu, Dong Dou, Zhicheng |
| author_facet | Deng, Chenlong Zhang, Zhisong Mao, Kelong Li, Shuaiyi Huang, Xinting Yu, Dong Dou, Zhicheng |
| contents | In this work, we provide a thorough investigation of gist-based context compression methods to improve long-context processing in large language models. We focus on two key questions: (1) How well can these methods replace full attention models? and (2) What potential failure patterns arise due to compression? Through extensive experiments, we show that while gist-based compression can achieve near-lossless performance on tasks like retrieval-augmented generation and long-document QA, it faces challenges in tasks like synthetic recall. Furthermore, we identify three key failure patterns: lost by the boundary, lost if surprise, and lost along the way. To mitigate these issues, we propose two effective strategies: fine-grained autoencoding, which enhances the reconstruction of original token information, and segment-wise token importance estimation, which adjusts optimization based on token dependencies. Our work provides valuable insights into the understanding of gist token-based context compression and offers practical strategies for improving compression capabilities. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_17483 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | A Silver Bullet or a Compromise for Full Attention? A Comprehensive Study of Gist Token-based Context Compression Deng, Chenlong Zhang, Zhisong Mao, Kelong Li, Shuaiyi Huang, Xinting Yu, Dong Dou, Zhicheng Computation and Language In this work, we provide a thorough investigation of gist-based context compression methods to improve long-context processing in large language models. We focus on two key questions: (1) How well can these methods replace full attention models? and (2) What potential failure patterns arise due to compression? Through extensive experiments, we show that while gist-based compression can achieve near-lossless performance on tasks like retrieval-augmented generation and long-document QA, it faces challenges in tasks like synthetic recall. Furthermore, we identify three key failure patterns: lost by the boundary, lost if surprise, and lost along the way. To mitigate these issues, we propose two effective strategies: fine-grained autoencoding, which enhances the reconstruction of original token information, and segment-wise token importance estimation, which adjusts optimization based on token dependencies. Our work provides valuable insights into the understanding of gist token-based context compression and offers practical strategies for improving compression capabilities. |
| title | A Silver Bullet or a Compromise for Full Attention? A Comprehensive Study of Gist Token-based Context Compression |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2412.17483 |