metadata

license: apache-2.0
tags:
  - generated_from_trainer
model-index:
  - name: bart-base-spelling-de
    results: []

bart-base-spelling-de

This model is a fine-tuned version of facebook/bart-base on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.1065
Cer: 0.2022

Model description

This is a proof of concept spelling correction model for german.

Intended uses & limitations

This is work in progress, be aware that the model can produce artifacts. You can test the model using the pipeline interface:

from transformers import pipeline

fix_spelling = pipeline("text2text-generation",model="oliverguhr/spelling-correction-german-base")

print(fix_spelling("das idst ein neuZr test",max_length=2048))

Training and evaluation data

More information needed

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
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.2803	0.11	1000	0.1978	0.9429
0.1688	0.21	2000	0.1472	0.9426
0.121	0.32	3000	0.1381	0.9424
0.1722	0.43	4000	0.1340	0.9425
0.1502	0.54	5000	0.1292	0.9423
0.1556	0.64	6000	0.1260	0.9424
0.1624	0.75	7000	0.1246	0.9425
0.1337	0.86	8000	0.1213	0.9424
0.131	0.96	9000	0.1195	0.9423
0.1137	1.07	10000	0.1178	0.9424
0.0958	1.18	11000	0.1166	0.9422
0.1067	1.28	12000	0.1147	0.9422
0.1201	1.39	13000	0.1135	0.9423
0.1115	1.5	14000	0.1111	0.9423
0.1284	1.61	15000	0.1101	0.9422
0.0947	1.71	16000	0.1085	0.9422
0.1081	1.82	17000	0.1073	0.9422
0.099	1.93	18000	0.1065	0.9422

Framework versions

Transformers 4.19.0.dev0
Pytorch 1.11.0+cu102
Datasets 2.1.0
Tokenizers 0.12.1