Papers with position-aware objective function
Landmark Embedding: A Chunking-Free Embedding Method For Retrieval Augmented Long-Context Large Language Models (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods for retrieval augmentation work with chunked contexts, which leads to poor quality of semantic representation and incomplete retrieval of useful information. |
| Approach: | They propose a method for retrieval augmentation of long-context language modeling using landmark embedding. |
| Outcome: | The proposed method outperforms existing retrieval methods with a notable advantage. |