File size: 4,772 Bytes
accd595
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
62
63
64
65
66
67
68
69
70
71
72
73
74
75
---
language: nl
license: mit
tags:
- flair
- token-classification
- sequence-tagger-model
base_model: dbmdz/bert-base-historic-multilingual-64k-td-cased
widget:
- text: Professoren der Geneeskun dige Faculteit te Groningen alsook van de HH , Doctoren
    en Chirurgijns van Groningen , Friesland , Noordholland , Overijssel , Gelderland
    , Drenthe , in welke Provinciën dit Elixir als Medicament voor Mond en Tanden
    reeds jaren bakend is .
---

# Fine-tuned Flair Model on Dutch ICDAR-Europeana NER Dataset

This Flair model was fine-tuned on the
[Dutch ICDAR-Europeana](https://github.com/stefan-it/historic-domain-adaptation-icdar)
NER Dataset using hmBERT 64k as backbone LM.

The ICDAR-Europeana NER Dataset is a preprocessed variant of the
[Europeana NER Corpora](https://github.com/EuropeanaNewspapers/ner-corpora) for Dutch and French.

The following NEs were annotated: `PER`, `LOC` and `ORG`.

# Results

We performed a hyper-parameter search over the following parameters with 5 different seeds per configuration:

* Batch Sizes: `[4, 8]`
* Learning Rates: `[3e-05, 5e-05]`

And report micro F1-score on development set:

| Configuration     | Seed 1          | Seed 2       | Seed 3       | Seed 4       | Seed 5       | Average         |
|-------------------|-----------------|--------------|--------------|--------------|--------------|-----------------|
| `bs8-e10-lr3e-05` | [0.8405][1]     | [0.8318][2]  | [0.8437][3]  | [0.8346][4]  | [0.8444][5]  | 0.839 ± 0.0056  |
| `bs4-e10-lr3e-05` | [**0.8467**][6] | [0.8303][7]  | [0.8238][8]  | [0.8386][9]  | [0.8274][10] | 0.8334 ± 0.0092 |
| `bs8-e10-lr5e-05` | [0.8284][11]    | [0.8345][12] | [0.831][13]  | [0.8229][14] | [0.8368][15] | 0.8307 ± 0.0054 |
| `bs4-e10-lr5e-05` | [0.8158][16]    | [0.8142][17] | [0.8164][18] | [0.8249][19] | [0.8228][20] | 0.8188 ± 0.0047 |

[1]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-1
[2]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-2
[3]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-3
[4]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-4
[5]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-5
[6]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-1
[7]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-2
[8]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-3
[9]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-4
[10]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-5
[11]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-1
[12]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-2
[13]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-3
[14]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-4
[15]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-5
[16]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-1
[17]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-2
[18]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-3
[19]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-4
[20]: https://hf.co/stefan-it/hmbench-icdar-nl-hmbert_64k-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-5

The [training log](training.log) and TensorBoard logs (not available for hmBERT Base model) are also uploaded to the model hub.

More information about fine-tuning can be found [here](https://github.com/stefan-it/hmBench).

# Acknowledgements

We thank [Luisa März](https://github.com/LuisaMaerz), [Katharina Schmid](https://github.com/schmika) and
[Erion Çano](https://github.com/erionc) for their fruitful discussions about Historic Language Models.

Research supported with Cloud TPUs from Google's [TPU Research Cloud](https://sites.research.google/trc/about/) (TRC).
Many Thanks for providing access to the TPUs ❤️