File size: 13,356 Bytes
0d51201
ab03fb1
 
 
 
15d2e3e
ab03fb1
0d51201
ab03fb1
15d2e3e
e7a5ee7
050e67d
 
 
ec49ce0
050e67d
15d2e3e
 
ab03fb1
 
 
050e67d
 
 
 
 
 
 
ab03fb1
 
 
050e67d
 
 
ab03fb1
 
 
050e67d
 
 
 
 
 
 
 
ab03fb1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c09032c
 
15d2e3e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ab03fb1
 
 
 
 
 
 
 
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
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
---
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: bart-base-spelling-nl-1m-3
  results: []
---

# bart-base-spelling-nl-1m-3

This model is a Dutch fine-tuned version of
[facebook/bart-base](https://huggingface.co/facebook/bart-base).

It achieves the following results on the evaluation set:

- Loss: 0.0053
- Cer: 0.0117

## Model description

This is a fine-tuned version of
[facebook/bart-base](https://huggingface.co/facebook/bart-base)
trained on spelling correction. It leans on the excellent work by
Oliver Guhr ([github](https://github.com/oliverguhr/spelling),
[huggingface](https://huggingface.co/oliverguhr/spelling-correction-english-base)). Training
was performed on an AWS EC2 instance (g5.xlarge) on a single GPU, and
took about two days.

## Intended uses & limitations

The intended use for this model is to be a component of the
[Valkuil.net](https://valkuil.net) context-sensitive spelling
checker. 

## Training and evaluation data

The model was trained on a Dutch dataset composed of 2,964,203 lines
of text from three public Dutch sources, downloaded from the
[Opus corpus](https://opus.nlpl.eu/):

- nl-europarlv7.1m.txt (1,000,000 lines)
- nl-opensubtitles2016.1m.txt (1,000,000 lines)
- nl-wikipedia.txt (964,203 lines)


## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.0003
- train_batch_size: 2
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 16
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 2.0

### Training results

| Training Loss | Epoch | Step   | Validation Loss | Cer    |
|:-------------:|:-----:|:------:|:---------------:|:------:|
| 0.0834        | 0.01  | 1000   | 0.0603          | 0.9216 |
| 0.0566        | 0.02  | 2000   | 0.0467          | 0.9217 |
| 0.0534        | 0.03  | 3000   | 0.0436          | 0.9216 |
| 0.0461        | 0.04  | 4000   | 0.0392          | 0.9216 |
| 0.0416        | 0.05  | 5000   | 0.0354          | 0.9216 |
| 0.0433        | 0.06  | 6000   | 0.0336          | 0.9216 |
| 0.045         | 0.08  | 7000   | 0.0315          | 0.9216 |
| 0.0452        | 0.09  | 8000   | 0.0305          | 0.9216 |
| 0.04          | 0.1   | 9000   | 0.0281          | 0.9216 |
| 0.0307        | 0.11  | 10000  | 0.0273          | 0.9216 |
| 0.0382        | 0.12  | 11000  | 0.0269          | 0.9216 |
| 0.036         | 0.13  | 12000  | 0.0254          | 0.9216 |
| 0.0412        | 0.14  | 13000  | 0.0258          | 0.9216 |
| 0.0404        | 0.15  | 14000  | 0.0238          | 0.9216 |
| 0.0265        | 0.16  | 15000  | 0.0239          | 0.9216 |
| 0.029         | 0.17  | 16000  | 0.0235          | 0.9216 |
| 0.0295        | 0.18  | 17000  | 0.0218          | 0.9216 |
| 0.0262        | 0.19  | 18000  | 0.0214          | 0.9216 |
| 0.0274        | 0.21  | 19000  | 0.0222          | 0.9216 |
| 0.0317        | 0.22  | 20000  | 0.0204          | 0.9216 |
| 0.0248        | 0.23  | 21000  | 0.0204          | 0.9216 |
| 0.0258        | 0.24  | 22000  | 0.0195          | 0.9216 |
| 0.0247        | 0.25  | 23000  | 0.0188          | 0.9216 |
| 0.0285        | 0.26  | 24000  | 0.0191          | 0.9215 |
| 0.031         | 0.27  | 25000  | 0.0192          | 0.9216 |
| 0.0267        | 0.28  | 26000  | 0.0188          | 0.9216 |
| 0.0245        | 0.29  | 27000  | 0.0177          | 0.9216 |
| 0.0258        | 0.3   | 28000  | 0.0177          | 0.9216 |
| 0.0235        | 0.31  | 29000  | 0.0169          | 0.9216 |
| 0.0235        | 0.32  | 30000  | 0.0176          | 0.9216 |
| 0.0223        | 0.34  | 31000  | 0.0165          | 0.9216 |
| 0.0219        | 0.35  | 32000  | 0.0167          | 0.9216 |
| 0.0214        | 0.36  | 33000  | 0.0165          | 0.9216 |
| 0.0232        | 0.37  | 34000  | 0.0163          | 0.9216 |
| 0.0192        | 0.38  | 35000  | 0.0162          | 0.9216 |
| 0.0159        | 0.39  | 36000  | 0.0160          | 0.9216 |
| 0.0205        | 0.4   | 37000  | 0.0150          | 0.9216 |
| 0.0197        | 0.41  | 38000  | 0.0152          | 0.9216 |
| 0.0205        | 0.42  | 39000  | 0.0150          | 0.9216 |
| 0.0182        | 0.43  | 40000  | 0.0145          | 0.9216 |
| 0.0204        | 0.44  | 41000  | 0.0139          | 0.9216 |
| 0.0201        | 0.45  | 42000  | 0.0146          | 0.9216 |
| 0.0202        | 0.46  | 43000  | 0.0132          | 0.9215 |
| 0.0219        | 0.48  | 44000  | 0.0146          | 0.9216 |
| 0.0161        | 0.49  | 45000  | 0.0134          | 0.9215 |
| 0.0172        | 0.5   | 46000  | 0.0137          | 0.9216 |
| 0.0199        | 0.51  | 47000  | 0.0133          | 0.9215 |
| 0.0211        | 0.52  | 48000  | 0.0132          | 0.9215 |
| 0.0184        | 0.53  | 49000  | 0.0136          | 0.9216 |
| 0.0191        | 0.54  | 50000  | 0.0129          | 0.9216 |
| 0.017         | 0.55  | 51000  | 0.0127          | 0.9216 |
| 0.0188        | 0.56  | 52000  | 0.0127          | 0.9215 |
| 0.0157        | 0.57  | 53000  | 0.0128          | 0.9216 |
| 0.0158        | 0.58  | 54000  | 0.0127          | 0.9216 |
| 0.0145        | 0.59  | 55000  | 0.0117          | 0.9216 |
| 0.0148        | 0.61  | 56000  | 0.0123          | 0.9216 |
| 0.0153        | 0.62  | 57000  | 0.0115          | 0.9216 |
| 0.0182        | 0.63  | 58000  | 0.0115          | 0.9216 |
| 0.0178        | 0.64  | 59000  | 0.0112          | 0.9215 |
| 0.0187        | 0.65  | 60000  | 0.0113          | 0.9215 |
| 0.0174        | 0.66  | 61000  | 0.0119          | 0.9216 |
| 0.0135        | 0.67  | 62000  | 0.0115          | 0.9215 |
| 0.0167        | 0.68  | 63000  | 0.0112          | 0.9216 |
| 0.0163        | 0.69  | 64000  | 0.0111          | 0.9215 |
| 0.0128        | 0.7   | 65000  | 0.0110          | 0.9215 |
| 0.0178        | 0.71  | 66000  | 0.0113          | 0.9215 |
| 0.0142        | 0.72  | 67000  | 0.0110          | 0.9215 |
| 0.0143        | 0.74  | 68000  | 0.0110          | 0.9215 |
| 0.0168        | 0.75  | 69000  | 0.0106          | 0.9216 |
| 0.0136        | 0.76  | 70000  | 0.0107          | 0.9215 |
| 0.0141        | 0.77  | 71000  | 0.0104          | 0.9215 |
| 0.0217        | 0.78  | 72000  | 0.0115          | 0.9216 |
| 0.012         | 0.79  | 73000  | 0.0105          | 0.9215 |
| 0.0141        | 0.8   | 74000  | 0.0100          | 0.9215 |
| 0.0136        | 0.81  | 75000  | 0.0096          | 0.9215 |
| 0.0106        | 0.82  | 76000  | 0.0104          | 0.9216 |
| 0.0176        | 0.83  | 77000  | 0.0102          | 0.9216 |
| 0.0169        | 0.84  | 78000  | 0.0099          | 0.9215 |
| 0.0118        | 0.85  | 79000  | 0.0102          | 0.9215 |
| 0.0178        | 0.86  | 80000  | 0.0095          | 0.9215 |
| 0.0145        | 0.88  | 81000  | 0.0097          | 0.9216 |
| 0.0154        | 0.89  | 82000  | 0.0099          | 0.9215 |
| 0.0129        | 0.9   | 83000  | 0.0094          | 0.9215 |
| 0.0125        | 0.91  | 84000  | 0.0097          | 0.9215 |
| 0.0147        | 0.92  | 85000  | 0.0093          | 0.9215 |
| 0.0145        | 0.93  | 86000  | 0.0091          | 0.9215 |
| 0.0121        | 0.94  | 87000  | 0.0089          | 0.9215 |
| 0.0125        | 0.95  | 88000  | 0.0094          | 0.9215 |
| 0.0113        | 0.96  | 89000  | 0.0088          | 0.9216 |
| 0.0098        | 0.97  | 90000  | 0.0094          | 0.9216 |
| 0.0137        | 0.98  | 91000  | 0.0089          | 0.9215 |
| 0.0105        | 0.99  | 92000  | 0.0091          | 0.9215 |
| 0.01          | 1.01  | 93000  | 0.0090          | 0.9216 |
| 0.0103        | 1.02  | 94000  | 0.0087          | 0.9216 |
| 0.0103        | 1.03  | 95000  | 0.0091          | 0.9215 |
| 0.0107        | 1.04  | 96000  | 0.0088          | 0.9216 |
| 0.0109        | 1.05  | 97000  | 0.0087          | 0.9215 |
| 0.0102        | 1.06  | 98000  | 0.0090          | 0.9216 |
| 0.0109        | 1.07  | 99000  | 0.0087          | 0.9215 |
| 0.0094        | 1.08  | 100000 | 0.0084          | 0.9215 |
| 0.009         | 1.09  | 101000 | 0.0085          | 0.9215 |
| 0.0085        | 1.1   | 102000 | 0.0084          | 0.9216 |
| 0.0123        | 1.11  | 103000 | 0.0085          | 0.9215 |
| 0.0094        | 1.12  | 104000 | 0.0084          | 0.9215 |
| 0.0076        | 1.14  | 105000 | 0.0081          | 0.9215 |
| 0.0119        | 1.15  | 106000 | 0.0079          | 0.9216 |
| 0.0079        | 1.16  | 107000 | 0.0081          | 0.9216 |
| 0.0108        | 1.17  | 108000 | 0.0080          | 0.9216 |
| 0.01          | 1.18  | 109000 | 0.0077          | 0.9216 |
| 0.0112        | 1.19  | 110000 | 0.0077          | 0.9216 |
| 0.0092        | 1.2   | 111000 | 0.0076          | 0.9215 |
| 0.0097        | 1.21  | 112000 | 0.0077          | 0.9215 |
| 0.0093        | 1.22  | 113000 | 0.0078          | 0.9215 |
| 0.0106        | 1.23  | 114000 | 0.0077          | 0.9215 |
| 0.0107        | 1.24  | 115000 | 0.0076          | 0.9215 |
| 0.0111        | 1.25  | 116000 | 0.0077          | 0.9215 |
| 0.0118        | 1.26  | 117000 | 0.0076          | 0.9215 |
| 0.0088        | 1.28  | 118000 | 0.0076          | 0.9215 |
| 0.01          | 1.29  | 119000 | 0.0076          | 0.9215 |
| 0.0102        | 1.3   | 120000 | 0.0076          | 0.9215 |
| 0.0106        | 1.31  | 121000 | 0.0076          | 0.9215 |
| 0.0099        | 1.32  | 122000 | 0.0077          | 0.9215 |
| 0.0099        | 1.33  | 123000 | 0.0077          | 0.9216 |
| 0.0105        | 1.34  | 124000 | 0.0075          | 0.9216 |
| 0.0082        | 1.35  | 125000 | 0.0074          | 0.9216 |
| 0.0088        | 1.36  | 126000 | 0.0072          | 0.9215 |
| 0.0077        | 1.37  | 127000 | 0.0070          | 0.9215 |
| 0.0063        | 1.38  | 128000 | 0.0074          | 0.9216 |
| 0.0084        | 1.39  | 129000 | 0.0069          | 0.9215 |
| 0.0085        | 1.41  | 130000 | 0.0071          | 0.9215 |
| 0.0107        | 1.42  | 131000 | 0.0067          | 0.9215 |
| 0.0064        | 1.43  | 132000 | 0.0068          | 0.9215 |
| 0.0064        | 1.44  | 133000 | 0.0069          | 0.9215 |
| 0.0139        | 1.45  | 134000 | 0.0067          | 0.9216 |
| 0.0093        | 1.46  | 135000 | 0.0068          | 0.9216 |
| 0.009         | 1.47  | 136000 | 0.0067          | 0.9215 |
| 0.0083        | 1.48  | 137000 | 0.0065          | 0.9216 |
| 0.0108        | 1.49  | 138000 | 0.0064          | 0.9215 |
| 0.0074        | 1.5   | 139000 | 0.0066          | 0.9215 |
| 0.009         | 1.51  | 140000 | 0.0064          | 0.9216 |
| 0.0062        | 1.52  | 141000 | 0.0064          | 0.9215 |
| 0.007         | 1.54  | 142000 | 0.0063          | 0.9215 |
| 0.0082        | 1.55  | 143000 | 0.0062          | 0.9215 |
| 0.0077        | 1.56  | 144000 | 0.0064          | 0.9215 |
| 0.0094        | 1.57  | 145000 | 0.0062          | 0.9215 |
| 0.0085        | 1.58  | 146000 | 0.0063          | 0.9215 |
| 0.0091        | 1.59  | 147000 | 0.0062          | 0.9215 |
| 0.0087        | 1.6   | 148000 | 0.0061          | 0.9215 |
| 0.0066        | 1.61  | 149000 | 0.0062          | 0.9215 |
| 0.0087        | 1.62  | 150000 | 0.0061          | 0.9215 |
| 0.0059        | 1.63  | 151000 | 0.0059          | 0.9215 |
| 0.0086        | 1.64  | 152000 | 0.0059          | 0.9215 |
| 0.0066        | 1.65  | 153000 | 0.0059          | 0.9215 |
| 0.0076        | 1.66  | 154000 | 0.0058          | 0.9215 |
| 0.0073        | 1.68  | 155000 | 0.0060          | 0.9215 |
| 0.0118        | 1.69  | 156000 | 0.0060          | 0.9215 |
| 0.0058        | 1.7   | 157000 | 0.0059          | 0.9215 |
| 0.0093        | 1.71  | 158000 | 0.0058          | 0.9215 |
| 0.0079        | 1.72  | 159000 | 0.0058          | 0.9215 |
| 0.0063        | 1.73  | 160000 | 0.0059          | 0.9215 |
| 0.0065        | 1.74  | 161000 | 0.0056          | 0.9215 |
| 0.0105        | 1.75  | 162000 | 0.0057          | 0.9215 |
| 0.0075        | 1.76  | 163000 | 0.0055          | 0.9215 |
| 0.0069        | 1.77  | 164000 | 0.0056          | 0.9215 |
| 0.0075        | 1.78  | 165000 | 0.0056          | 0.9215 |
| 0.0067        | 1.79  | 166000 | 0.0055          | 0.9215 |
| 0.0069        | 1.81  | 167000 | 0.0056          | 0.9215 |
| 0.0063        | 1.82  | 168000 | 0.0056          | 0.9215 |
| 0.0058        | 1.83  | 169000 | 0.0055          | 0.9215 |
| 0.0058        | 1.84  | 170000 | 0.0054          | 0.9215 |
| 0.0081        | 1.85  | 171000 | 0.0055          | 0.9215 |
| 0.0071        | 1.86  | 172000 | 0.0054          | 0.9215 |
| 0.0077        | 1.87  | 173000 | 0.0054          | 0.9215 |
| 0.0053        | 1.88  | 174000 | 0.0053          | 0.9215 |
| 0.0067        | 1.89  | 175000 | 0.0053          | 0.9215 |
| 0.0066        | 1.9   | 176000 | 0.0053          | 0.9215 |
| 0.0084        | 1.91  | 177000 | 0.0053          | 0.9215 |
| 0.0066        | 1.92  | 178000 | 0.0052          | 0.9215 |
| 0.0057        | 1.94  | 179000 | 0.0053          | 0.9215 |
| 0.0059        | 1.95  | 180000 | 0.0052          | 0.9215 |
| 0.0053        | 1.96  | 181000 | 0.0053          | 0.9215 |
| 0.0056        | 1.97  | 182000 | 0.0052          | 0.9215 |
| 0.0054        | 1.98  | 183000 | 0.0052          | 0.9215 |
| 0.0053        | 1.99  | 184000 | 0.0052          | 0.9215 |
| 0.0066        | 2.0   | 185000 | 0.0052          | 0.9215 |


### Framework versions

- Transformers 4.27.3
- Pytorch 2.0.0+cu117
- Datasets 2.10.1
- Tokenizers 0.13.2