distilbert-base-uncased-imdb

This model is a fine-tuned version of distilbert-base-uncased on an imdb dataset where an evaluation of 5000 samples was created by splitting the training set. It achieves the following results on the evaluation set:

• Loss: 0.6252
• Accuracy: 0.9214

Model description

More information needed

Intended uses & limitations

This model was trained for the introduction to Natural language processing course of EPITA.

Training and evaluation data

The training/evaluation split was generated using a seed of 42 and a test_size of 0.2.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 32
• eval_batch_size: 128
• seed: 1337
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.2875	1.0	625	0.2286	0.9102
0.1685	2.0	1250	0.2416	0.9128
0.1171	3.0	1875	0.3223	0.917
0.0493	4.0	2500	0.3667	0.9162
0.023	5.0	3125	0.4074	0.92
0.015	6.0	3750	0.4291	0.9236
0.0129	7.0	4375	0.5452	0.9194
0.0051	8.0	5000	0.5886	0.9146
0.0027	9.0	5625	0.6310	0.9186
0.002	10.0	6250	0.6252	0.9214

Framework versions

• Transformers 4.20.1
• Pytorch 1.11.0+cu113
• Datasets 2.3.2
• Tokenizers 0.12.1