RichardErkhov
commited on
Commit
•
39f970a
1
Parent(s):
2321ffb
uploaded readme
Browse files
README.md
ADDED
@@ -0,0 +1,163 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
Quantization made by Richard Erkhov.
|
2 |
+
|
3 |
+
[Github](https://github.com/RichardErkhov)
|
4 |
+
|
5 |
+
[Discord](https://discord.gg/pvy7H8DZMG)
|
6 |
+
|
7 |
+
[Request more models](https://github.com/RichardErkhov/quant_request)
|
8 |
+
|
9 |
+
|
10 |
+
xlm-roberta-base-squad2 - bnb 8bits
|
11 |
+
- Model creator: https://huggingface.co/deepset/
|
12 |
+
- Original model: https://huggingface.co/deepset/xlm-roberta-base-squad2/
|
13 |
+
|
14 |
+
|
15 |
+
|
16 |
+
|
17 |
+
Original model description:
|
18 |
+
---
|
19 |
+
license: cc-by-4.0
|
20 |
+
datasets:
|
21 |
+
- squad_v2
|
22 |
+
model-index:
|
23 |
+
- name: deepset/xlm-roberta-base-squad2
|
24 |
+
results:
|
25 |
+
- task:
|
26 |
+
type: question-answering
|
27 |
+
name: Question Answering
|
28 |
+
dataset:
|
29 |
+
name: squad_v2
|
30 |
+
type: squad_v2
|
31 |
+
config: squad_v2
|
32 |
+
split: validation
|
33 |
+
metrics:
|
34 |
+
- type: exact_match
|
35 |
+
value: 74.0354
|
36 |
+
name: Exact Match
|
37 |
+
verified: true
|
38 |
+
verifyToken: eyJhbGciOiJFZERTQSIsInR5cCI6IkpXVCJ9.eyJoYXNoIjoiMWMxNWQ2ODJkNWIzZGQwOWI4OTZjYjU3ZDVjZGQzMjI5MzljNjliZTY4Mzk4YTk4OTMzZWYxZjUxYmZhYTBhZSIsInZlcnNpb24iOjF9.eEeFYYJ30BfJDd-JYfI1kjlxJrRF6OFtj2GnkTCOO4kqX31inFy8ptDWusVlLFsUphm4dNWfTKXC5e-gytLBDA
|
39 |
+
- type: f1
|
40 |
+
value: 77.1833
|
41 |
+
name: F1
|
42 |
+
verified: true
|
43 |
+
verifyToken: eyJhbGciOiJFZERTQSIsInR5cCI6IkpXVCJ9.eyJoYXNoIjoiMjg4MjNkOTA4Y2I5OGFlYTk1NWZjMWFlNjI5M2Y0NGZhMThhN2M4YmY2Y2RhZjcwYzU0MGNjN2RkZDljZmJmNiIsInZlcnNpb24iOjF9.TX42YMXpH4e0qu7cC4ARDlZWSkd55dwwyeyFXmOlXERNnEicDuFBCsy8WHLaqQCLUkzODJ22Hw4zhv81rwnlAQ
|
44 |
+
---
|
45 |
+
|
46 |
+
# Multilingual XLM-RoBERTa base for QA on various languages
|
47 |
+
|
48 |
+
## Overview
|
49 |
+
**Language model:** xlm-roberta-base
|
50 |
+
**Language:** Multilingual
|
51 |
+
**Downstream-task:** Extractive QA
|
52 |
+
**Training data:** SQuAD 2.0
|
53 |
+
**Eval data:** SQuAD 2.0 dev set - German MLQA - German XQuAD
|
54 |
+
**Code:** See [example](https://github.com/deepset-ai/FARM/blob/master/examples/question_answering.py) in [FARM](https://github.com/deepset-ai/FARM/blob/master/examples/question_answering.py)
|
55 |
+
**Infrastructure**: 4x Tesla v100
|
56 |
+
|
57 |
+
## Hyperparameters
|
58 |
+
|
59 |
+
```
|
60 |
+
batch_size = 22*4
|
61 |
+
n_epochs = 2
|
62 |
+
max_seq_len=256,
|
63 |
+
doc_stride=128,
|
64 |
+
learning_rate=2e-5,
|
65 |
+
```
|
66 |
+
|
67 |
+
Corresponding experiment logs in mlflow: [link](https://public-mlflow.deepset.ai/#/experiments/2/runs/b25ec75e07614accb3f1ce03d43dbe08)
|
68 |
+
|
69 |
+
|
70 |
+
## Performance
|
71 |
+
Evaluated on the SQuAD 2.0 dev set with the [official eval script](https://worksheets.codalab.org/rest/bundles/0x6b567e1cf2e041ec80d7098f031c5c9e/contents/blob/).
|
72 |
+
```
|
73 |
+
"exact": 73.91560683904657
|
74 |
+
"f1": 77.14103746689592
|
75 |
+
```
|
76 |
+
|
77 |
+
Evaluated on German MLQA: test-context-de-question-de.json
|
78 |
+
"exact": 33.67279167589108
|
79 |
+
"f1": 44.34437105434842
|
80 |
+
"total": 4517
|
81 |
+
|
82 |
+
Evaluated on German XQuAD: xquad.de.json
|
83 |
+
"exact": 48.739495798319325
|
84 |
+
"f1": 62.552615701071495
|
85 |
+
"total": 1190
|
86 |
+
|
87 |
+
|
88 |
+
## Usage
|
89 |
+
|
90 |
+
### In Transformers
|
91 |
+
```python
|
92 |
+
from transformers.pipelines import pipeline
|
93 |
+
from transformers.modeling_auto import AutoModelForQuestionAnswering
|
94 |
+
from transformers.tokenization_auto import AutoTokenizer
|
95 |
+
|
96 |
+
model_name = "deepset/xlm-roberta-base-squad2"
|
97 |
+
|
98 |
+
# a) Get predictions
|
99 |
+
nlp = pipeline('question-answering', model=model_name, tokenizer=model_name)
|
100 |
+
QA_input = {
|
101 |
+
'question': 'Why is model conversion important?',
|
102 |
+
'context': 'The option to convert models between FARM and transformers gives freedom to the user and let people easily switch between frameworks.'
|
103 |
+
}
|
104 |
+
res = nlp(QA_input)
|
105 |
+
|
106 |
+
# b) Load model & tokenizer
|
107 |
+
model = AutoModelForQuestionAnswering.from_pretrained(model_name)
|
108 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
109 |
+
```
|
110 |
+
|
111 |
+
### In FARM
|
112 |
+
|
113 |
+
```python
|
114 |
+
from farm.modeling.adaptive_model import AdaptiveModel
|
115 |
+
from farm.modeling.tokenization import Tokenizer
|
116 |
+
from farm.infer import Inferencer
|
117 |
+
|
118 |
+
model_name = "deepset/xlm-roberta-base-squad2"
|
119 |
+
|
120 |
+
# a) Get predictions
|
121 |
+
nlp = Inferencer.load(model_name, task_type="question_answering")
|
122 |
+
QA_input = [{"questions": ["Why is model conversion important?"],
|
123 |
+
"text": "The option to convert models between FARM and transformers gives freedom to the user and let people easily switch between frameworks."}]
|
124 |
+
res = nlp.inference_from_dicts(dicts=QA_input, rest_api_schema=True)
|
125 |
+
|
126 |
+
# b) Load model & tokenizer
|
127 |
+
model = AdaptiveModel.convert_from_transformers(model_name, device="cpu", task_type="question_answering")
|
128 |
+
tokenizer = Tokenizer.load(model_name)
|
129 |
+
```
|
130 |
+
|
131 |
+
### In haystack
|
132 |
+
For doing QA at scale (i.e. many docs instead of single paragraph), you can load the model also in [haystack](https://github.com/deepset-ai/haystack/):
|
133 |
+
```python
|
134 |
+
reader = FARMReader(model_name_or_path="deepset/xlm-roberta-base-squad2")
|
135 |
+
# or
|
136 |
+
reader = TransformersReader(model="deepset/roberta-base-squad2",tokenizer="deepset/xlm-roberta-base-squad2")
|
137 |
+
```
|
138 |
+
|
139 |
+
|
140 |
+
## Authors
|
141 |
+
Branden Chan: `branden.chan [at] deepset.ai`
|
142 |
+
Timo Möller: `timo.moeller [at] deepset.ai`
|
143 |
+
Malte Pietsch: `malte.pietsch [at] deepset.ai`
|
144 |
+
Tanay Soni: `tanay.soni [at] deepset.ai`
|
145 |
+
|
146 |
+
## About us
|
147 |
+
![deepset logo](https://workablehr.s3.amazonaws.com/uploads/account/logo/476306/logo)
|
148 |
+
|
149 |
+
We bring NLP to the industry via open source!
|
150 |
+
Our focus: Industry specific language models & large scale QA systems.
|
151 |
+
|
152 |
+
Some of our work:
|
153 |
+
- [German BERT (aka "bert-base-german-cased")](https://deepset.ai/german-bert)
|
154 |
+
- [GermanQuAD and GermanDPR datasets and models (aka "gelectra-base-germanquad", "gbert-base-germandpr")](https://deepset.ai/germanquad)
|
155 |
+
- [FARM](https://github.com/deepset-ai/FARM)
|
156 |
+
- [Haystack](https://github.com/deepset-ai/haystack/)
|
157 |
+
|
158 |
+
Get in touch:
|
159 |
+
[Twitter](https://twitter.com/deepset_ai) | [LinkedIn](https://www.linkedin.com/company/deepset-ai/) | [Discord](https://haystack.deepset.ai/community/join) | [GitHub Discussions](https://github.com/deepset-ai/haystack/discussions) | [Website](https://deepset.ai)
|
160 |
+
|
161 |
+
By the way: [we're hiring!](http://www.deepset.ai/jobs)
|
162 |
+
|
163 |
+
|