mobilevit-xx-small-v2024-10-22

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

• Loss: 0.1725
• Accuracy: 0.9338
• F1: 0.8269
• Precision: 0.8260
• Recall: 0.8279

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: 0.0002
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.6549	1.7544	100	0.6289	0.82	0.6260	0.5191	0.7884
0.4616	3.5088	200	0.4192	0.8867	0.7296	0.6706	0.8
0.3101	5.2632	300	0.3071	0.9036	0.7318	0.7810	0.6884
0.2932	7.0175	400	0.2486	0.908	0.7460	0.7896	0.7070
0.2652	8.7719	500	0.2279	0.9138	0.7674	0.7921	0.7442
0.2253	10.5263	600	0.2100	0.9218	0.7859	0.8240	0.7512
0.2257	12.2807	700	0.1951	0.9249	0.8019	0.8085	0.7953
0.2468	14.0351	800	0.1906	0.9307	0.8199	0.8142	0.8256
0.1796	15.7895	900	0.1949	0.9276	0.8120	0.8055	0.8186
0.1888	17.5439	1000	0.1807	0.9307	0.8178	0.8216	0.8140
0.202	19.2982	1100	0.1772	0.9342	0.8287	0.8249	0.8326
0.1824	21.0526	1200	0.1826	0.9276	0.8080	0.8186	0.7977
0.1808	22.8070	1300	0.1682	0.9347	0.8297	0.8268	0.8326
0.1792	24.5614	1400	0.1688	0.9364	0.8324	0.8392	0.8256
0.1852	26.3158	1500	0.1725	0.9338	0.8269	0.8260	0.8279
0.177	28.0702	1600	0.1690	0.9351	0.8282	0.8381	0.8186
0.1857	29.8246	1700	0.1708	0.9298	0.8176	0.8119	0.8233

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.2
• Tokenizers 0.19.1