--- 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