stefan-it commited on
Commit
4fb6a85
1 Parent(s): f526174

readme: add initial version (#1)

Browse files

- readme: add initial version (3d95bfaeabaf3c411b1b2e82be1983253efc397f)

Files changed (1) hide show
  1. README.md +85 -0
README.md ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: en
3
+ license: mit
4
+ tags:
5
+ - flair
6
+ - token-classification
7
+ - sequence-tagger-model
8
+ base_model: hmbyt5-preliminary/byt5-small-historic-multilingual-span20-flax
9
+ inference: false
10
+ widget:
11
+ - text: On Wednesday , a public dinner was given by the Conservative Burgesses of
12
+ Leads , to the Conservative members of the Leeds Town Council , in the Music Hall
13
+ , Albion-street , which was very numerously attended .
14
+ ---
15
+
16
+ # Fine-tuned Flair Model on TopRes19th English NER Dataset (HIPE-2022)
17
+
18
+ This Flair model was fine-tuned on the
19
+ [TopRes19th English](https://github.com/hipe-eval/HIPE-2022-data/blob/main/documentation/README-topres19th.md)
20
+ NER Dataset using hmByT5 as backbone LM.
21
+
22
+ The TopRes19th dataset consists of NE-annotated historical English newspaper articles from 19C.
23
+
24
+ The following NEs were annotated: `BUILDING`, `LOC` and `STREET`.
25
+
26
+ # ⚠️ Inference Widget ⚠️
27
+
28
+ Fine-Tuning ByT5 models in Flair is currently done by implementing an own [`ByT5Embedding`][1] class.
29
+
30
+ This class needs to be present when running the model with Flair.
31
+
32
+ Thus, the inference widget is not working with hmByT5 at the moment on the Model Hub and is currently disabled.
33
+
34
+ This should be fixed in future, when ByT5 fine-tuning is supported in Flair directly.
35
+
36
+ [1]: https://github.com/stefan-it/hmBench/blob/main/byt5_embeddings.py
37
+
38
+ # Results
39
+
40
+ We performed a hyper-parameter search over the following parameters with 5 different seeds per configuration:
41
+
42
+ * Batch Sizes: `[8, 4]`
43
+ * Learning Rates: `[0.00015, 0.00016]`
44
+
45
+ And report micro F1-score on development set:
46
+
47
+ | Configuration | Run 1 | Run 2 | Run 3 | Run 4 | Run 5 | Avg. |
48
+ |-------------------|--------------|--------------|--------------|--------------|--------------|--------------|
49
+ | bs4-e10-lr0.00015 | [0.7992][1] | [0.8226][2] | [0.8205][3] | [0.8364][4] | [0.809][5] | 81.75 ± 1.26 |
50
+ | bs8-e10-lr0.00015 | [0.8095][6] | [0.83][7] | [0.8024][8] | [0.8112][9] | [0.8189][10] | 81.44 ± 0.94 |
51
+ | bs8-e10-lr0.00016 | [0.8144][11] | [0.8209][12] | [0.8065][13] | [0.8056][14] | [0.82][15] | 81.35 ± 0.65 |
52
+ | bs4-e10-lr0.00016 | [0.8056][16] | [0.8105][17] | [0.809][18] | [0.808][19] | [0.8056][20] | 80.77 ± 0.19 |
53
+
54
+ [1]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs4-wsFalse-e10-lr0.00015-poolingfirst-layers-1-crfFalse-1
55
+ [2]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs4-wsFalse-e10-lr0.00015-poolingfirst-layers-1-crfFalse-2
56
+ [3]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs4-wsFalse-e10-lr0.00015-poolingfirst-layers-1-crfFalse-3
57
+ [4]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs4-wsFalse-e10-lr0.00015-poolingfirst-layers-1-crfFalse-4
58
+ [5]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs4-wsFalse-e10-lr0.00015-poolingfirst-layers-1-crfFalse-5
59
+ [6]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs8-wsFalse-e10-lr0.00015-poolingfirst-layers-1-crfFalse-1
60
+ [7]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs8-wsFalse-e10-lr0.00015-poolingfirst-layers-1-crfFalse-2
61
+ [8]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs8-wsFalse-e10-lr0.00015-poolingfirst-layers-1-crfFalse-3
62
+ [9]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs8-wsFalse-e10-lr0.00015-poolingfirst-layers-1-crfFalse-4
63
+ [10]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs8-wsFalse-e10-lr0.00015-poolingfirst-layers-1-crfFalse-5
64
+ [11]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs8-wsFalse-e10-lr0.00016-poolingfirst-layers-1-crfFalse-1
65
+ [12]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs8-wsFalse-e10-lr0.00016-poolingfirst-layers-1-crfFalse-2
66
+ [13]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs8-wsFalse-e10-lr0.00016-poolingfirst-layers-1-crfFalse-3
67
+ [14]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs8-wsFalse-e10-lr0.00016-poolingfirst-layers-1-crfFalse-4
68
+ [15]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs8-wsFalse-e10-lr0.00016-poolingfirst-layers-1-crfFalse-5
69
+ [16]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs4-wsFalse-e10-lr0.00016-poolingfirst-layers-1-crfFalse-1
70
+ [17]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs4-wsFalse-e10-lr0.00016-poolingfirst-layers-1-crfFalse-2
71
+ [18]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs4-wsFalse-e10-lr0.00016-poolingfirst-layers-1-crfFalse-3
72
+ [19]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs4-wsFalse-e10-lr0.00016-poolingfirst-layers-1-crfFalse-4
73
+ [20]: https://hf.co/hmbench/hmbench-topres19th-en-hmbyt5-bs4-wsFalse-e10-lr0.00016-poolingfirst-layers-1-crfFalse-5
74
+
75
+ The [training log](training.log) and TensorBoard logs are also uploaded to the model hub.
76
+
77
+ More information about fine-tuning can be found [here](https://github.com/stefan-it/hmBench).
78
+
79
+ # Acknowledgements
80
+
81
+ We thank [Luisa März](https://github.com/LuisaMaerz), [Katharina Schmid](https://github.com/schmika) and
82
+ [Erion Çano](https://github.com/erionc) for their fruitful discussions about Historic Language Models.
83
+
84
+ Research supported with Cloud TPUs from Google's [TPU Research Cloud](https://sites.research.google/trc/about/) (TRC).
85
+ Many Thanks for providing access to the TPUs ❤️