mobilevitv2-1.0-imagenet1k-256-finetuned_v2024-7-25-frost

This model is a fine-tuned version of apple/mobilevitv2-1.0-imagenet1k-256 on the webdataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.1896
• Accuracy: 0.9310
• F1: 0.8227
• Precision: 0.8458
• Recall: 0.8009

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
0.6687	1.5625	100	0.6623	0.7230	0.5335	0.4022	0.7920
0.4454	3.125	200	0.4152	0.8832	0.7490	0.6567	0.8717
0.2835	4.6875	300	0.2661	0.9097	0.7661	0.7952	0.7389
0.2197	6.25	400	0.2151	0.9195	0.7869	0.8358	0.7434
0.1613	7.8125	500	0.2007	0.9292	0.8140	0.8578	0.7743
0.1655	9.375	600	0.1935	0.9310	0.8227	0.8458	0.8009
0.1815	10.9375	700	0.1883	0.9265	0.8074	0.8488	0.7699
0.1316	12.5	800	0.1825	0.9327	0.8273	0.8505	0.8053
0.1612	14.0625	900	0.1837	0.9257	0.8100	0.8287	0.7920
0.118	15.625	1000	0.1896	0.9310	0.8227	0.8458	0.8009
0.1178	17.1875	1100	0.1937	0.9239	0.8028	0.8333	0.7743
0.1248	18.75	1200	0.1913	0.9301	0.8192	0.8483	0.7920
0.1169	20.3125	1300	0.1916	0.9301	0.8167	0.8585	0.7788
0.1094	21.875	1400	0.1925	0.9292	0.8182	0.8411	0.7965
0.1108	23.4375	1500	0.1961	0.9345	0.8333	0.8486	0.8186
0.1089	25.0	1600	0.1993	0.9283	0.8172	0.8341	0.8009
0.0919	26.5625	1700	0.1936	0.9319	0.8262	0.8433	0.8097
0.0969	28.125	1800	0.1978	0.9310	0.8227	0.8458	0.8009
0.1093	29.6875	1900	0.1955	0.9283	0.8172	0.8341	0.8009

Framework versions

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