File size: 1,356 Bytes
957690c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
afaec19
957690c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e52197d
957690c
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
---
task_categories:
- token-classification
language:
- en
tags:
- hallucination
- rag
pretty_name: RAGTruth++
---
# Dataset Card for RAGTruth++

This dataset is a re-annotation of a subset of the RAGTruth-benchmark that can be used to evaluate hallucination detection models.

It contains 408 prompt-response-pairs and 865 spans that are marked as hallucinations.

A text span is marked as a hallucination when a generated output in the response is not supported by the data in the input prompt.

You can read more about the creation of the dataset on [our blog](https://www.blueguardrails.com/en/blog/ragtruth-plus-plus-enhanced-hallucination-detection-benchmark).

## Structure of the dataset

We provide two CSV-files.

One with hallucination spans, following the following format:

```json
{
  "start_pos": "<int>",
  "end_pos": "<int>",
  "span_text": "<str>",
  "label": "<str>",
  "reasoning": "<str>",
  "stable_id": "<str>",
  "message_stable_id": "<str>"  
}
```

One with messages:

```json
{
  "stable_id": "<str>",
  "meta": "<str>",
  "role": "<user|assistant>",
  "position": "<int>",
  "text": "<str>",
  "message_thread_stable_id": "<str>"
}
```

Use the `message_stable_id`-field to reference spans to messages.


## Uses

Use the dataset for benchmarking hallucination detection models.

## Dataset Card Contact

info@blueguardrails.com