Papers with LongFaith
LongFaith: Enhancing Long-Context Reasoning in LLMs with Faithful Synthetic Data (2025.findings-acl)
Copied to clipboard
| Challenge: | Long-context processing ability has emerged as a significant challenge for large language models. |
| Approach: | They propose a pipeline for synthesizing faithful long-context reasoning instruction datasets . they integrate ground truth and citation-based reasoning prompts integrating them . |
| Outcome: | The proposed pipeline eliminates distractions and improves reasoning chains. |