readme: add new DistilBERT-related sections
Browse files
README.md
CHANGED
@@ -4,3 +4,51 @@ license: mit
|
|
4 |
tags:
|
5 |
- "historic german"
|
6 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
tags:
|
5 |
- "historic german"
|
6 |
---
|
7 |
+
|
8 |
+
# π€ + π dbmdz DistilBERT model
|
9 |
+
|
10 |
+
In this repository the MDZ Digital Library team (dbmdz) at the Bavarian State
|
11 |
+
Library open sources a German Europeana DistilBERT model π
|
12 |
+
|
13 |
+
# German Europeana DistilBERT
|
14 |
+
|
15 |
+
We use the open source [Europeana newspapers](http://www.europeana-newspapers.eu/)
|
16 |
+
that were provided by *The European Library*. The final
|
17 |
+
training corpus has a size of 51GB and consists of 8,035,986,369 tokens.
|
18 |
+
|
19 |
+
Detailed information about the data and pretraining steps can be found in
|
20 |
+
[this repository](https://github.com/stefan-it/europeana-bert).
|
21 |
+
|
22 |
+
## Results
|
23 |
+
|
24 |
+
For results on Historic NER, please refer to [this repository](https://github.com/stefan-it/europeana-bert).
|
25 |
+
|
26 |
+
## Usage
|
27 |
+
|
28 |
+
With Transformers >= 4.3 our German Europeana DistilBERT model can be loaded like:
|
29 |
+
|
30 |
+
```python
|
31 |
+
from transformers import AutoModel, AutoTokenizer
|
32 |
+
|
33 |
+
model_name = "distilbert-base-german-europeana-cased"
|
34 |
+
|
35 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
36 |
+
model = AutoModel.from_pretrained(model_name)
|
37 |
+
```
|
38 |
+
|
39 |
+
# Huggingface model hub
|
40 |
+
|
41 |
+
All other German Europeana models are available on the [Huggingface model hub](https://huggingface.co/dbmdz).
|
42 |
+
|
43 |
+
# Contact (Bugs, Feedback, Contribution and more)
|
44 |
+
|
45 |
+
For questions about our Europeana BERT, ELECTRA and ConvBERT models just open a new discussion
|
46 |
+
[here](https://github.com/stefan-it/europeana-bert/discussions) π€
|
47 |
+
|
48 |
+
# Acknowledgments
|
49 |
+
|
50 |
+
Research supported with Cloud TPUs from Google's TensorFlow Research Cloud (TFRC).
|
51 |
+
Thanks for providing access to the TFRC β€οΈ
|
52 |
+
|
53 |
+
Thanks to the generous support from the [Hugging Face](https://huggingface.co/) team,
|
54 |
+
it is possible to download both cased and uncased models from their S3 storage π€
|