levit_128.fb_dist_in1k-finetuned-stroke-binary

This model is a fine-tuned version of timm/levit_128.fb_dist_in1k on an binary stroke detection dataset. It achieves the following results on the evaluation set:

• Accuracy: 0.8598
• F1: 0.8577
• Precision: 0.8602
• Recall: 0.8598

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: 36
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.7002	0.6202	100	nan	0.5690	0.5387	0.5349	0.5690
0.681	1.2357	200	nan	0.5834	0.5331	0.5372	0.5834
0.6874	1.8558	300	nan	0.6002	0.5596	0.5665	0.6002
0.6774	2.4713	400	nan	0.6124	0.5811	0.5867	0.6124
0.6533	3.0868	500	nan	0.6852	0.6694	0.6767	0.6852
0.6368	3.7070	600	nan	0.7205	0.7153	0.7153	0.7205
0.6196	4.3225	700	nan	0.7603	0.7471	0.7650	0.7603
0.5663	4.9426	800	nan	0.7883	0.7843	0.7864	0.7883
0.5196	5.5581	900	nan	0.8078	0.7972	0.8206	0.8078
0.4704	6.1736	1000	nan	0.8363	0.8317	0.8396	0.8363
0.4715	6.7938	1100	nan	0.8349	0.8292	0.8409	0.8349
0.452	7.4093	1200	nan	0.8503	0.8479	0.8505	0.8503
0.4538	8.0248	1300	nan	0.8598	0.8577	0.8602	0.8598

Framework versions

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