rohanphanse
commited on
Commit
·
2ad544b
1
Parent(s):
7052635
Update README.md
Browse files
README.md
CHANGED
|
@@ -26,4 +26,47 @@ configs:
|
|
| 26 |
- split: corpus
|
| 27 |
path: story/corpus.jsonl
|
| 28 |
---
|
| 29 |
-
# MSRS: Evaluating Multi-Source Retrieval-Augmented Generation
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 26 |
- split: corpus
|
| 27 |
path: story/corpus.jsonl
|
| 28 |
---
|
| 29 |
+
# MSRS: Evaluating Multi-Source Retrieval-Augmented Generation
|
| 30 |
+
|
| 31 |
+
#### [**📄 Paper**](https://arxiv.org/abs/2508.20867) | [**💻 Code**](https://github.com/yale-nlp/MSRS)
|
| 32 |
+
|
| 33 |
+
This paper introduces a scalable framework for constructing evaluation benchmarks that challenge RAG systems to integrate information across distinct sources and generate long-form responses. Using our framework, we build two new benchmarks on Multi-Source Retrieval and Synthesis: MSRS-Story and MSRS-Meet.
|
| 34 |
+
|
| 35 |
+
## 🚀 Quickstart
|
| 36 |
+
|
| 37 |
+
Load the corpora for MSRS-Story and MSRS-Meet:
|
| 38 |
+
```py
|
| 39 |
+
from datasets import load_dataset
|
| 40 |
+
|
| 41 |
+
story_corpus = load_dataset("yale-nlp/MSRS", "story-corpus", split="corpus")
|
| 42 |
+
meeting_corpus = load_dataset("yale-nlp/MSRS", "meeting-corpus", split="corpus")
|
| 43 |
+
```
|
| 44 |
+
|
| 45 |
+
Corpus Dataset Example:
|
| 46 |
+
|
| 47 |
+
```js
|
| 48 |
+
{
|
| 49 |
+
"id": // Unique ID for the document
|
| 50 |
+
"text": // Document text
|
| 51 |
+
}
|
| 52 |
+
```
|
| 53 |
+
|
| 54 |
+
Load the query-answer pairs for MSRS-Story and MSRS-Meet (available splits: `train`, `test`, and `validation`):
|
| 55 |
+
|
| 56 |
+
```py
|
| 57 |
+
from datasets import load_dataset
|
| 58 |
+
|
| 59 |
+
story_qa = load_dataset("yale-nlp/MSRS", "story-qa")
|
| 60 |
+
meeting_qa = load_dataset("yale-nlp/MSRS", "meeting-qa")
|
| 61 |
+
```
|
| 62 |
+
|
| 63 |
+
QA Dataset Example:
|
| 64 |
+
|
| 65 |
+
```js
|
| 66 |
+
{
|
| 67 |
+
"id": // Unique ID for the query
|
| 68 |
+
"query": // Query text
|
| 69 |
+
"gold_documents": // List of gold document IDs
|
| 70 |
+
"answer": // List of answer summaries
|
| 71 |
+
}
|
| 72 |
+
```
|