File size: 7,596 Bytes
0d51201
ab03fb1
 
 
 
 
 
0d51201
ab03fb1
 
 
2912c8b
 
ab03fb1
 
2912c8b
 
ab03fb1
 
 
 
2912c8b
 
 
 
 
 
ab03fb1
 
 
2912c8b
 
 
ab03fb1
 
 
2912c8b
 
 
ab03fb1
2912c8b
 
 
ab03fb1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2912c8b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: bart-base-spelling-nl
  results: []
---

# bart-base-spelling-nl

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: 0.0217
- Cer: 0.0147

## Model description

This is a text-to-text 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.

## 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. A next version of the model will be trained on more data.

## Training and evaluation data

The model was trained on a Dutch dataset composed of 1,500,000 lines of
text from three public Dutch sources, downloaded from the [Opus
corpus](https://opus.nlpl.eu/):

- nl-europarlv7.100k.txt (500,000 lines)
- nl-opensubtitles2016.100k.txt (500,000 lines)
- nl-wikipedia.100k.txt (500,000 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    |
|:-------------:|:-----:|:-----:|:---------------:|:------:|
| 0.2546        | 0.02  | 1000  | 0.1801          | 0.9245 |
| 0.1646        | 0.04  | 2000  | 0.1203          | 0.9243 |
| 0.1456        | 0.06  | 3000  | 0.1016          | 0.9242 |
| 0.1204        | 0.09  | 4000  | 0.0849          | 0.9242 |
| 0.1226        | 0.11  | 5000  | 0.0736          | 0.9241 |
| 0.1049        | 0.13  | 6000  | 0.0680          | 0.9240 |
| 0.1071        | 0.15  | 7000  | 0.0671          | 0.9241 |
| 0.1038        | 0.17  | 8000  | 0.0615          | 0.9240 |
| 0.0815        | 0.19  | 9000  | 0.0575          | 0.9240 |
| 0.0828        | 0.21  | 10000 | 0.0572          | 0.9241 |
| 0.0851        | 0.24  | 11000 | 0.0533          | 0.9241 |
| 0.0787        | 0.26  | 12000 | 0.0529          | 0.9241 |
| 0.0795        | 0.28  | 13000 | 0.0518          | 0.9239 |
| 0.0864        | 0.3   | 14000 | 0.0492          | 0.9239 |
| 0.0806        | 0.32  | 15000 | 0.0471          | 0.9239 |
| 0.0808        | 0.34  | 16000 | 0.0483          | 0.9238 |
| 0.071         | 0.36  | 17000 | 0.0469          | 0.9239 |
| 0.0661        | 0.38  | 18000 | 0.0446          | 0.9239 |
| 0.0641        | 0.41  | 19000 | 0.0437          | 0.9239 |
| 0.0686        | 0.43  | 20000 | 0.0428          | 0.9238 |
| 0.0597        | 0.45  | 21000 | 0.0431          | 0.9238 |
| 0.0585        | 0.47  | 22000 | 0.0417          | 0.9238 |
| 0.0675        | 0.49  | 23000 | 0.0406          | 0.9238 |
| 0.0678        | 0.51  | 24000 | 0.0395          | 0.9238 |
| 0.0581        | 0.53  | 25000 | 0.0393          | 0.9238 |
| 0.0569        | 0.56  | 26000 | 0.0371          | 0.9239 |
| 0.0632        | 0.58  | 27000 | 0.0378          | 0.9238 |
| 0.0589        | 0.6   | 28000 | 0.0377          | 0.9238 |
| 0.0511        | 0.62  | 29000 | 0.0366          | 0.9237 |
| 0.0651        | 0.64  | 30000 | 0.0358          | 0.9239 |
| 0.0594        | 0.66  | 31000 | 0.0356          | 0.9238 |
| 0.054         | 0.68  | 32000 | 0.0368          | 0.9238 |
| 0.0498        | 0.71  | 33000 | 0.0353          | 0.9238 |
| 0.0559        | 0.73  | 34000 | 0.0337          | 0.9238 |
| 0.0502        | 0.75  | 35000 | 0.0341          | 0.9238 |
| 0.0588        | 0.77  | 36000 | 0.0339          | 0.9239 |
| 0.0487        | 0.79  | 37000 | 0.0338          | 0.9237 |
| 0.0489        | 0.81  | 38000 | 0.0333          | 0.9236 |
| 0.0493        | 0.83  | 39000 | 0.0331          | 0.9237 |
| 0.0481        | 0.85  | 40000 | 0.0323          | 0.9237 |
| 0.0444        | 0.88  | 41000 | 0.0318          | 0.9237 |
| 0.0446        | 0.9   | 42000 | 0.0311          | 0.9238 |
| 0.0469        | 0.92  | 43000 | 0.0311          | 0.9237 |
| 0.0525        | 0.94  | 44000 | 0.0312          | 0.9237 |
| 0.042         | 0.96  | 45000 | 0.0312          | 0.9236 |
| 0.0541        | 0.98  | 46000 | 0.0304          | 0.9237 |
| 0.0417        | 1.0   | 47000 | 0.0293          | 0.9238 |
| 0.0369        | 1.03  | 48000 | 0.0305          | 0.9237 |
| 0.0357        | 1.05  | 49000 | 0.0297          | 0.9237 |
| 0.0394        | 1.07  | 50000 | 0.0296          | 0.9237 |
| 0.0343        | 1.09  | 51000 | 0.0288          | 0.9237 |
| 0.037         | 1.11  | 52000 | 0.0286          | 0.9237 |
| 0.0367        | 1.13  | 53000 | 0.0281          | 0.9237 |
| 0.0336        | 1.15  | 54000 | 0.0287          | 0.9236 |
| 0.0331        | 1.18  | 55000 | 0.0275          | 0.9237 |
| 0.0381        | 1.2   | 56000 | 0.0276          | 0.9237 |
| 0.0438        | 1.22  | 57000 | 0.0269          | 0.9237 |
| 0.0319        | 1.24  | 58000 | 0.0274          | 0.9236 |
| 0.0364        | 1.26  | 59000 | 0.0265          | 0.9237 |
| 0.0402        | 1.28  | 60000 | 0.0262          | 0.9237 |
| 0.0341        | 1.3   | 61000 | 0.0259          | 0.9237 |
| 0.0346        | 1.32  | 62000 | 0.0258          | 0.9237 |
| 0.0378        | 1.35  | 63000 | 0.0258          | 0.9236 |
| 0.0372        | 1.37  | 64000 | 0.0253          | 0.9237 |
| 0.0375        | 1.39  | 65000 | 0.0248          | 0.9237 |
| 0.0336        | 1.41  | 66000 | 0.0246          | 0.9236 |
| 0.031         | 1.43  | 67000 | 0.0246          | 0.9237 |
| 0.0344        | 1.45  | 68000 | 0.0248          | 0.9236 |
| 0.0307        | 1.47  | 69000 | 0.0244          | 0.9236 |
| 0.0293        | 1.5   | 70000 | 0.0239          | 0.9237 |
| 0.0406        | 1.52  | 71000 | 0.0235          | 0.9236 |
| 0.0273        | 1.54  | 72000 | 0.0235          | 0.9236 |
| 0.0316        | 1.56  | 73000 | 0.0234          | 0.9235 |
| 0.0308        | 1.58  | 74000 | 0.0229          | 0.9236 |
| 0.0291        | 1.6   | 75000 | 0.0229          | 0.9236 |
| 0.0325        | 1.62  | 76000 | 0.0229          | 0.9236 |
| 0.0347        | 1.65  | 77000 | 0.0224          | 0.9237 |
| 0.0268        | 1.67  | 78000 | 0.0226          | 0.9237 |
| 0.0279        | 1.69  | 79000 | 0.0219          | 0.9236 |
| 0.0247        | 1.71  | 80000 | 0.0220          | 0.9235 |
| 0.0259        | 1.73  | 81000 | 0.0215          | 0.9236 |
| 0.0294        | 1.75  | 82000 | 0.0217          | 0.9235 |
| 0.0267        | 1.77  | 83000 | 0.0217          | 0.9236 |
| 0.0273        | 1.79  | 84000 | 0.0213          | 0.9236 |
| 0.0242        | 1.82  | 85000 | 0.0213          | 0.9236 |
| 0.0254        | 1.84  | 86000 | 0.0210          | 0.9236 |
| 0.0273        | 1.86  | 87000 | 0.0209          | 0.9236 |
| 0.0261        | 1.88  | 88000 | 0.0210          | 0.9235 |
| 0.0244        | 1.9   | 89000 | 0.0206          | 0.9235 |
| 0.0256        | 1.92  | 90000 | 0.0206          | 0.9235 |
| 0.0283        | 1.94  | 91000 | 0.0205          | 0.9235 |
| 0.0255        | 1.97  | 92000 | 0.0204          | 0.9235 |
| 0.022         | 1.99  | 93000 | 0.0203          | 0.9235 |


### Framework versions

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