deit-base-patch16-224-finetuned-stroke-binary

This model is a fine-tuned version of facebook/deit-base-patch16-224 on an BTX24/tekno21-brain-stroke-dataset-binary dataset. It achieves the following results on the evaluation set:

• Loss: 0.1527
• Accuracy: 0.9489
• F1: 0.9484
• Precision: 0.9505
• Recall: 0.9489

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 48
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.1646	0.6202	100	0.1588	0.9430	0.9425	0.9442	0.9430
0.1417	1.2357	200	0.1640	0.9439	0.9433	0.9458	0.9439
0.1681	1.8558	300	0.1622	0.9453	0.9447	0.9470	0.9453
0.1512	2.4713	400	0.1510	0.9435	0.9430	0.9441	0.9435
0.1506	3.0868	500	0.1913	0.9340	0.9327	0.9391	0.9340
0.1654	3.7070	600	0.1679	0.9426	0.9419	0.9442	0.9426
0.1482	4.3225	700	0.1551	0.9403	0.9402	0.9402	0.9403
0.1599	4.9426	800	0.1489	0.9462	0.9457	0.9471	0.9462
0.1477	5.5581	900	0.1437	0.9426	0.9424	0.9425	0.9426
0.1308	6.1736	1000	0.1527	0.9417	0.9414	0.9416	0.9417
0.1362	6.7938	1100	0.1608	0.9426	0.9421	0.9432	0.9426
0.1494	7.4093	1200	0.1601	0.9435	0.9429	0.9451	0.9435
0.1592	8.0248	1300	0.1430	0.9430	0.9429	0.9429	0.9430
0.16	8.6450	1400	0.1504	0.9457	0.9451	0.9475	0.9457
0.1245	9.2605	1500	0.1506	0.9462	0.9458	0.9470	0.9462
0.1397	9.8806	1600	0.1971	0.9313	0.9300	0.9359	0.9313
0.1396	10.4961	1700	0.1527	0.9489	0.9484	0.9505	0.9489

Framework versions

• Transformers 4.48.3
• Pytorch 2.6.0+cu124
• Datasets 3.4.0
• Tokenizers 0.21.0