Commit
•
f009860
1
Parent(s):
6a0289d
Upload Readme
Browse files
README.md
CHANGED
|
@@ -96,20 +96,7 @@ configs:
|
|
| 96 |
---
|
| 97 |
|
| 98 |
# L-CITEEVAL: DO LONG-CONTEXT MODELS TRULY LEVERAGE CONTEXT FOR RESPONDING?
|
| 99 |
-
<p
|
| 100 |
-
<strong>Paper</strong>
|
| 101 |
-
<a href="https://arxiv.org/abs/2410.02115">
|
| 102 |
-
<img src="https://img.shields.io/badge/arXiv-2410.02115-b31b1b.svg?style=plastic" style="vertical-align: middle;">
|
| 103 |
-
</a>
|
| 104 |
-
<strong>Github</strong>
|
| 105 |
-
<a href="https://github.com/ZetangForward/L-CITEEVAL">
|
| 106 |
-
<img src="https://github.githubassets.com/assets/GitHub-Mark-ea2971cee799.png" alt="Github" width="40" height="40" style="vertical-align: middle;">
|
| 107 |
-
</a>
|
| 108 |
-
<strong>Zhihu</strong>
|
| 109 |
-
<a href="https://zhuanlan.zhihu.com/p/817442176">
|
| 110 |
-
<img src="https://img.shields.io/badge/知乎-0079FF.svg?style=plastic&logo=zhihu&logoColor=white" style="vertical-align: middle;">
|
| 111 |
-
</a>
|
| 112 |
-
</p>
|
| 113 |
|
| 114 |
## Benchmark Quickview
|
| 115 |
*L-CiteEval* is a multi-task long-context understanding with citation benchmark, covering **5 task categories**, including single-document question answering, multi-document question answering, summarization, dialogue understanding, and synthetic tasks, encompassing **11 different long-context tasks**. The context lengths for these tasks range from **8K to 48K**.
|
|
@@ -144,7 +131,7 @@ All data in L-CiteEval follows the format below:
|
|
| 144 |
"answer": "The correct or expected answer to the question, used for evaluating correctness",
|
| 145 |
"docs": "Context divided into fixed-length chunks"
|
| 146 |
"length": "The context length"
|
| 147 |
-
"hardness": "The level of
|
| 148 |
}
|
| 149 |
```
|
| 150 |
|
|
|
|
| 96 |
---
|
| 97 |
|
| 98 |
# L-CITEEVAL: DO LONG-CONTEXT MODELS TRULY LEVERAGE CONTEXT FOR RESPONDING?
|
| 99 |
+
**Paper** [](https://arxiv.org/abs/2410.02115) **Github** <a href="https://github.com/ZetangForward/L-CITEEVAL"><img src="https://github.githubassets.com/assets/GitHub-Mark-ea2971cee799.png" alt="Github" width="40" height="40"></a> **Zhihu** [](https://zhuanlan.zhihu.com/p/817442176)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
|
| 101 |
## Benchmark Quickview
|
| 102 |
*L-CiteEval* is a multi-task long-context understanding with citation benchmark, covering **5 task categories**, including single-document question answering, multi-document question answering, summarization, dialogue understanding, and synthetic tasks, encompassing **11 different long-context tasks**. The context lengths for these tasks range from **8K to 48K**.
|
|
|
|
| 131 |
"answer": "The correct or expected answer to the question, used for evaluating correctness",
|
| 132 |
"docs": "Context divided into fixed-length chunks"
|
| 133 |
"length": "The context length"
|
| 134 |
+
"hardness": "The level of difficulty in L-CiteEval-Hardness, which can be easy, medium and hard"
|
| 135 |
}
|
| 136 |
```
|
| 137 |
|