File size: 4,858 Bytes
90a450d |
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: fr
license: mit
tags:
- flair
- token-classification
- sequence-tagger-model
base_model: dbmdz/bert-tiny-historic-multilingual-cased
widget:
- text: Le Moniteur universel fait ressortir les avantages de la situation de l '
Allemagne , sa force militaire , le peu d ' intérêts personnels qu ' elle peut
avoir dans la question d ' Orient .
---
# Fine-tuned Flair Model on French NewsEye NER Dataset (HIPE-2022)
This Flair model was fine-tuned on the
[French NewsEye](https://github.com/hipe-eval/HIPE-2022-data/blob/main/documentation/README-newseye.md)
NER Dataset using hmBERT Tiny as backbone LM.
The NewsEye dataset is comprised of diachronic historical newspaper material published between 1850 and 1950
in French, German, Finnish, and Swedish.
More information can be found [here](https://dl.acm.org/doi/abs/10.1145/3404835.3463255).
The following NEs were annotated: `PER`, `LOC`, `ORG` and `HumanProd`.
# Results
We performed a hyper-parameter search over the following parameters with 5 different seeds per configuration:
* Batch Sizes: `[4, 8]`
* Learning Rates: `[5e-05, 3e-05]`
And report micro F1-score on development set:
| Configuration | Seed 1 | Seed 2 | Seed 3 | Seed 4 | Seed 5 | Average |
|-------------------|--------------|-----------------|--------------|--------------|--------------|-----------------|
| `bs4-e10-lr5e-05` | [0.5782][1] | [**0.5584**][2] | [0.5555][3] | [0.5685][4] | [0.5422][5] | 0.5606 ± 0.0136 |
| `bs8-e10-lr5e-05` | [0.5486][6] | [0.5273][7] | [0.5282][8] | [0.5288][9] | [0.5067][10] | 0.5279 ± 0.0148 |
| `bs4-e10-lr3e-05` | [0.5251][11] | [0.5103][12] | [0.5041][13] | [0.5124][14] | [0.479][15] | 0.5062 ± 0.017 |
| `bs8-e10-lr3e-05` | [0.4815][16] | [0.4879][17] | [0.4783][18] | [0.4648][19] | [0.4628][20] | 0.4751 ± 0.0109 |
[1]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-1
[2]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-2
[3]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-3
[4]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-4
[5]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-5
[6]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-1
[7]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-2
[8]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-3
[9]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-4
[10]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-5
[11]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-1
[12]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-2
[13]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-3
[14]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-4
[15]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-5
[16]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-1
[17]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-2
[18]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-3
[19]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-4
[20]: https://hf.co/stefan-it/hmbench-newseye-fr-hmbert_tiny-bs8-wsFalse-e10-lr3e-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 ❤️
|