mnist-mobilevit

This model is a fine-tuned version of apple/mobilevit-xx-small on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0185
• Accuracy: 0.9929

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: 0.0008
• train_batch_size: 128
• eval_batch_size: 128
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.026	0.2132	100	0.0458	0.9867
0.0232	0.4264	200	0.0453	0.9867
0.0277	0.6397	300	0.0484	0.9863
0.0293	0.8529	400	0.0469	0.9865
0.0235	1.0661	500	0.0288	0.9899
0.0203	1.2793	600	0.0253	0.9924
0.0182	1.4925	700	0.0286	0.9916
0.0205	1.7058	800	0.0203	0.9935
0.0162	1.9190	900	0.0238	0.9913
0.0118	2.1322	1000	0.0247	0.9916
0.0121	2.3454	1100	0.0194	0.9932
0.0154	2.5586	1200	0.0194	0.9933
0.015	2.7719	1300	0.0216	0.9933
0.0145	2.9851	1400	0.0238	0.9919
0.0098	3.1983	1500	0.0208	0.993
0.0093	3.4115	1600	0.0218	0.9929
0.0073	3.6247	1700	0.0189	0.9933
0.008	3.8380	1800	0.0194	0.9932
0.006	4.0512	1900	0.0183	0.9938
0.0063	4.2644	2000	0.0184	0.9934
0.0043	4.4776	2100	0.0184	0.9932
0.0035	4.6908	2200	0.0183	0.9931
0.0061	4.9041	2300	0.0184	0.9931

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1