README.md · xlm-roberta-base-squad2

1

---

2

license: cc-by-4.0

3

datasets:

4

- squad_v2

5

model-index:

6

- name: deepset/xlm-roberta-base-squad2

7

results:

8

- task:

9

type: question-answering

10

name: Question Answering

11

dataset:

12

name: squad_v2

13

type: squad_v2

14

config: squad_v2

15

split: validation

16

metrics:

17

- type: exact_match

18

value: 74.0354

19

name: Exact Match

20

verified: true

21

      verifyToken: eyJhbGciOiJFZERTQSIsInR5cCI6IkpXVCJ9.eyJoYXNoIjoiMWMxNWQ2ODJkNWIzZGQwOWI4OTZjYjU3ZDVjZGQzMjI5MzljNjliZTY4Mzk4YTk4OTMzZWYxZjUxYmZhYTBhZSIsInZlcnNpb24iOjF9.eEeFYYJ30BfJDd-JYfI1kjlxJrRF6OFtj2GnkTCOO4kqX31inFy8ptDWusVlLFsUphm4dNWfTKXC5e-gytLBDA

22

- type: f1

23

value: 77.1833

24

name: F1

25

verified: true

26

      verifyToken: eyJhbGciOiJFZERTQSIsInR5cCI6IkpXVCJ9.eyJoYXNoIjoiMjg4MjNkOTA4Y2I5OGFlYTk1NWZjMWFlNjI5M2Y0NGZhMThhN2M4YmY2Y2RhZjcwYzU0MGNjN2RkZDljZmJmNiIsInZlcnNpb24iOjF9.TX42YMXpH4e0qu7cC4ARDlZWSkd55dwwyeyFXmOlXERNnEicDuFBCsy8WHLaqQCLUkzODJ22Hw4zhv81rwnlAQ

27

---

28

29

# Multilingual XLM-RoBERTa base for Extractive QA on various languages

30

31

## Overview

32

**Language model:** xlm-roberta-base

33

**Language:** Multilingual

34

**Downstream-task:** Extractive QA

35

**Training data:** SQuAD 2.0

36

**Eval data:** SQuAD 2.0 dev set - German MLQA - German XQuAD

37

**Code:**  See [an example extractive QA pipeline built with Haystack](https://haystack.deepset.ai/tutorials/34_extractive_qa_pipeline)

38

**Infrastructure**: 4x Tesla v100

39

40

## Hyperparameters

41

42

```

43

batch_size = 22*4

44

n_epochs = 2

45

max_seq_len=256,

46

doc_stride=128,

47

learning_rate=2e-5,

48

```

49

50

Corresponding experiment logs in mlflow: [link](https://public-mlflow.deepset.ai/#/experiments/2/runs/b25ec75e07614accb3f1ce03d43dbe08)

51

52

53

## Usage

54

55

### In Haystack

56

Haystack is an AI orchestration framework to build customizable, production-ready LLM applications. You can use this model in Haystack to do extractive question answering on documents.

57

To load and run the model with [Haystack](https://github.com/deepset-ai/haystack/):

58

```python

59

# After running pip install haystack-ai "transformers[torch,sentencepiece]"

60

61

from haystack import Document

62

from haystack.components.readers import ExtractiveReader

63

64

docs = [

65

Document(content="Python is a popular programming language"),

66

Document(content="python ist eine beliebte Programmiersprache"),

67

]

68

69

reader = ExtractiveReader(model="deepset/xlm-roberta-base-squad2")

70

reader.warm_up()

71

72

question = "What is a popular programming language?"

73

result = reader.run(query=question, documents=docs)

74

# {'answers': [ExtractedAnswer(query='What is a popular programming language?', score=0.5740374326705933, data='python', document=Document(id=..., content: '...'), context=None, document_offset=ExtractedAnswer.Span(start=0, end=6),...)]}

75

```

76

For a complete example with an extractive question answering pipeline that scales over many documents, check out the [corresponding Haystack tutorial](https://haystack.deepset.ai/tutorials/34_extractive_qa_pipeline).

77

78

### In Transformers

79

```python

80

from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline

81

82

model_name = "deepset/xlm-roberta-base-squad2"

83

84

# a) Get predictions

85

nlp = pipeline('question-answering', model=model_name, tokenizer=model_name)

86

QA_input = {

87

'question': 'Why is model conversion important?',

88

    'context': 'The option to convert models between FARM and transformers gives freedom to the user and let people easily switch between frameworks.'

89

}

90

res = nlp(QA_input)

91

92

# b) Load model & tokenizer

93

model = AutoModelForQuestionAnswering.from_pretrained(model_name)

94

tokenizer = AutoTokenizer.from_pretrained(model_name)

95

```

96

97

## Performance

98

Evaluated on the SQuAD 2.0 dev set with the [official eval script](https://worksheets.codalab.org/rest/bundles/0x6b567e1cf2e041ec80d7098f031c5c9e/contents/blob/).

99

```

100

"exact": 73.91560683904657

101

"f1": 77.14103746689592

102

```

103

104

Evaluated on German MLQA: test-context-de-question-de.json

105

"exact": 33.67279167589108

106

"f1": 44.34437105434842

107

"total": 4517

108

109

Evaluated on German XQuAD: xquad.de.json

110

"exact": 48.739495798319325

111

"f1": 62.552615701071495

112

"total": 1190

113

114

## Authors

115

Branden Chan: `branden.chan [at] deepset.ai`

116

Timo Möller: `timo.moeller [at] deepset.ai`

117

Malte Pietsch: `malte.pietsch [at] deepset.ai`

118

Tanay Soni: `tanay.soni [at] deepset.ai`

119

120

## About us

121

122

<div class="grid lg:grid-cols-2 gap-x-4 gap-y-3">

123

<div class="w-full h-40 object-cover mb-2 rounded-lg flex items-center justify-center">

124

         <img alt="" src="https://raw.githubusercontent.com/deepset-ai/.github/main/deepset-logo-colored.png" class="w-40"/>

125

</div>

126

<div class="w-full h-40 object-cover mb-2 rounded-lg flex items-center justify-center">

127

         <img alt="" src="https://raw.githubusercontent.com/deepset-ai/.github/main/haystack-logo-colored.png" class="w-40"/>

128

</div>

129

</div>

130

131

[deepset](http://deepset.ai/) is the company behind the production-ready open-source AI framework [Haystack](https://haystack.deepset.ai/).

132

133

Some of our other work:

134

- [Distilled roberta-base-squad2 (aka "tinyroberta-squad2")](https://huggingface.co/deepset/tinyroberta-squad2)

135

- [German BERT](https://deepset.ai/german-bert), [GermanQuAD and GermanDPR](https://deepset.ai/germanquad), [German embedding model](https://huggingface.co/mixedbread-ai/deepset-mxbai-embed-de-large-v1)

136

- [deepset Cloud](https://www.deepset.ai/deepset-cloud-product), [deepset Studio](https://www.deepset.ai/deepset-studio)

137

138

## Get in touch and join the Haystack community

139

140

<p>For more info on Haystack, visit our <strong><a href="https://github.com/deepset-ai/haystack">GitHub</a></strong> repo and <strong><a href="https://docs.haystack.deepset.ai">Documentation</a></strong>.

141

142

We also have a <strong><a class="h-7" href="https://haystack.deepset.ai/community">Discord community open to everyone!</a></strong></p>

143

144

[Twitter](https://twitter.com/Haystack_AI) | [LinkedIn](https://www.linkedin.com/company/deepset-ai/) | [Discord](https://haystack.deepset.ai/community) | [GitHub Discussions](https://github.com/deepset-ai/haystack/discussions) | [Website](https://haystack.deepset.ai/) | [YouTube](https://www.youtube.com/@deepset_ai)

145

146

By the way: [we're hiring!](http://www.deepset.ai/jobs)