vit-emotions-fp16

This model is a fine-tuned version of google/vit-base-patch16-224-in21k on the imagefolder dataset. It achieves the following results on the evaluation set:

• Loss: 0.3051
• Accuracy: 0.9287

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: 5e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 25

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	50	1.7679	0.3862
No log	2.0	100	1.4584	0.5375
No log	3.0	150	1.3209	0.5162
No log	4.0	200	1.1580	0.62
No log	5.0	250	0.9946	0.7275
No log	6.0	300	0.8519	0.7887
No log	7.0	350	0.7374	0.8325
No log	8.0	400	0.7250	0.815
No log	9.0	450	0.5821	0.88
1.1152	10.0	500	0.5239	0.8838
1.1152	11.0	550	0.5121	0.8712
1.1152	12.0	600	0.4444	0.9038
1.1152	13.0	650	0.3894	0.9137
1.1152	14.0	700	0.3956	0.9137
1.1152	15.0	750	0.3806	0.91
1.1152	16.0	800	0.3328	0.9375
1.1152	17.0	850	0.3076	0.9287
1.1152	18.0	900	0.3026	0.9363
1.1152	19.0	950	0.2388	0.96
0.3752	20.0	1000	0.2892	0.935
0.3752	21.0	1050	0.2539	0.9413
0.3752	22.0	1100	0.2299	0.9525
0.3752	23.0	1150	0.2131	0.9575
0.3752	24.0	1200	0.2300	0.9525
0.3752	25.0	1250	0.2393	0.9537

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.17.0
• Tokenizers 0.15.1