stefan-it commited on
Commit
f7e5f0b
1 Parent(s): fd24f19

readme: add initial version

Browse files

This introduces the initial version of model card.

Files changed (1) hide show
  1. README.md +71 -0
README.md ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: fr
3
+ license: mit
4
+ tags:
5
+ - flair
6
+ - token-classification
7
+ - sequence-tagger-model
8
+ base_model: hmteams/teams-base-historic-multilingual-discriminator
9
+ widget:
10
+ - text: 'Parmi les remèdes recommandés par la Société , il faut mentionner celui que
11
+ M . Schatzmann , de Lausanne , a proposé :'
12
+ ---
13
+
14
+ # Fine-tuned Flair Model on LeTemps French NER Dataset (HIPE-2022)
15
+
16
+ This Flair model was fine-tuned on the
17
+ [LeTemps French](https://github.com/hipe-eval/HIPE-2022-data/blob/main/documentation/README-letemps.md)
18
+ NER Dataset using hmTEAMS as backbone LM.
19
+
20
+ The LeTemps dataset consists of NE-annotated historical French newspaper articles from mid-19C to mid 20C.
21
+
22
+ The following NEs were annotated: `loc`, `org` and `pers`.
23
+
24
+ # Results
25
+
26
+ We performed a hyper-parameter search over the following parameters with 5 different seeds per configuration:
27
+
28
+ * Batch Sizes: `[8, 4]`
29
+ * Learning Rates: `[3e-05, 5e-05]`
30
+
31
+ And report micro F1-score on development set:
32
+
33
+ | Configuration | Run 1 | Run 2 | Run 3 | Run 4 | Run 5 | Avg. |
34
+ |-----------------|--------------|--------------|--------------|--------------|--------------|--------------|
35
+ | bs8-e10-lr3e-05 | [0.6651][1] | [0.6542][2] | [0.66][3] | [0.6705][4] | [0.6702][5] | 66.4 ± 0.62 |
36
+ | bs4-e10-lr3e-05 | [0.66][6] | [0.6641][7] | [0.6641][8] | [0.6595][9] | [0.6548][10] | 66.05 ± 0.35 |
37
+ | bs8-e10-lr5e-05 | [0.6564][11] | [0.6555][12] | [0.6598][13] | [0.6581][14] | [0.6636][15] | 65.87 ± 0.29 |
38
+ | bs4-e10-lr5e-05 | [0.6415][16] | [0.6602][17] | [0.601][18] | [0.6505][19] | [0.6638][20] | 64.34 ± 2.26 |
39
+
40
+ [1]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-1
41
+ [2]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-2
42
+ [3]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-3
43
+ [4]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-4
44
+ [5]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs8-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-5
45
+ [6]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-1
46
+ [7]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-2
47
+ [8]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-3
48
+ [9]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-4
49
+ [10]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs4-wsFalse-e10-lr3e-05-poolingfirst-layers-1-crfFalse-5
50
+ [11]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-1
51
+ [12]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-2
52
+ [13]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-3
53
+ [14]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-4
54
+ [15]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs8-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-5
55
+ [16]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-1
56
+ [17]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-2
57
+ [18]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-3
58
+ [19]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-4
59
+ [20]: https://hf.co/stefan-it/hmbench-letemps-fr-hmteams-bs4-wsFalse-e10-lr5e-05-poolingfirst-layers-1-crfFalse-5
60
+
61
+ The [training log](training.log) and TensorBoard logs (only for hmByT5 and hmTEAMS based models) are also uploaded to the model hub.
62
+
63
+ More information about fine-tuning can be found [here](https://github.com/stefan-it/hmBench).
64
+
65
+ # Acknowledgements
66
+
67
+ We thank [Luisa März](https://github.com/LuisaMaerz), [Katharina Schmid](https://github.com/schmika) and
68
+ [Erion Çano](https://github.com/erionc) for their fruitful discussions about Historic Language Models.
69
+
70
+ Research supported with Cloud TPUs from Google's [TPU Research Cloud](https://sites.research.google/trc/about/) (TRC).
71
+ Many Thanks for providing access to the TPUs ❤️