mit-b2-VF2-finetuned-memes

This model is a fine-tuned version of nvidia/mit-b2 on the imagefolder dataset. It achieves the following results on the evaluation set:

• Loss: 0.6547
• Accuracy: 0.8308
• Precision: 0.8272
• Recall: 0.8308
• F1: 0.8287

Model description

More information needed

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: 0.00012
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1
1.3077	0.99	20	1.1683	0.5549	0.5621	0.5549	0.5286
0.9359	1.99	40	0.8573	0.6731	0.6807	0.6731	0.6535
0.7219	2.99	60	0.7106	0.7272	0.7359	0.7272	0.7246
0.6013	3.99	80	0.6445	0.7550	0.7686	0.7550	0.7558
0.5243	4.99	100	0.6717	0.7573	0.8077	0.7573	0.7584
0.4409	5.99	120	0.5315	0.8068	0.8027	0.8068	0.7989
0.3325	6.99	140	0.5159	0.8230	0.8236	0.8230	0.8158
0.2719	7.99	160	0.5250	0.8215	0.8227	0.8215	0.8202
0.242	8.99	180	0.5087	0.8277	0.8260	0.8277	0.8268
0.2247	9.99	200	0.5313	0.8215	0.8275	0.8215	0.8218
0.1955	10.99	220	0.6167	0.8130	0.8062	0.8130	0.8073
0.1567	11.99	240	0.5859	0.8168	0.8185	0.8168	0.8173
0.1479	12.99	260	0.5938	0.8215	0.8169	0.8215	0.8178
0.1241	13.99	280	0.6187	0.8261	0.8234	0.8261	0.8239
0.1114	14.99	300	0.6419	0.8261	0.8351	0.8261	0.8293
0.1022	15.99	320	0.6322	0.8323	0.8284	0.8323	0.8294
0.0941	16.99	340	0.6595	0.8269	0.8266	0.8269	0.8263
0.0935	17.99	360	0.6674	0.8269	0.8218	0.8269	0.8237
0.089	18.99	380	0.6533	0.8253	0.8222	0.8253	0.8235
0.0794	19.99	400	0.6547	0.8308	0.8272	0.8308	0.8287

Framework versions

• Transformers 4.24.0.dev0
• Pytorch 1.11.0+cu102
• Datasets 2.6.1.dev0
• Tokenizers 0.13.1