t5-end2end-questions-generation-dutch

This model is a fine-tuned version of yhavinga/t5-base-dutch on a Google translated version of SQUAD 1.1 found here: https://www.kaggle.com/datasets/michelvanheijningen/squad1-dutch.

The code used to finetune the model is largely based on the work by Thomas Simonini. You can find his English model here and his Google colab tutorial here

It achieves the following results on the evaluation set:

• Loss: 1.6546

Model description

This is my first model ever and still a work in progress ;)

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 7

Training results

Training Loss	Epoch	Step	Validation Loss
2.6528	0.34	100	1.9249
1.964	0.68	200	1.7897
1.8695	1.02	300	1.7554
1.7922	1.35	400	1.7270
1.7747	1.69	500	1.7054
1.7473	2.03	600	1.7019
1.697	2.37	700	1.6868
1.6848	2.71	800	1.6810
1.6756	3.05	900	1.6779
1.6282	3.39	1000	1.6712
1.6285	3.73	1100	1.6626
1.6161	4.06	1200	1.6616
1.5887	4.4	1300	1.6588
1.5877	4.74	1400	1.6583
1.5723	5.08	1500	1.6560
1.5545	5.42	1600	1.6550
1.5415	5.76	1700	1.6540
1.5509	6.1	1800	1.6541
1.5326	6.44	1900	1.6539
1.5268	6.77	2000	1.6546

Framework versions

• Transformers 4.27.4
• Pytorch 1.13.0
• Datasets 2.1.0
• Tokenizers 0.13.2