swin-base-food101-jpqd-ov

It was compressed using NNCF with Optimum Intel following the JPQD image classification example.

This model is a fine-tuned version of microsoft/swin-base-patch4-window7-224 on the food101 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3396
• Accuracy: 0.9061

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 16
• eval_batch_size: 128
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
2.2162	0.42	500	2.1111	0.7967
0.729	0.84	1000	0.5474	0.8773
0.7536	1.27	1500	0.3844	0.8984
0.4822	1.69	2000	0.3340	0.9043
12.2559	2.11	2500	12.0128	0.9033
48.7302	2.54	3000	48.3874	0.8681
75.1831	2.96	3500	75.3200	0.7183
93.5572	3.38	4000	93.4142	0.5939
103.798	3.8	4500	103.4427	0.5634
108.0993	4.23	5000	108.6461	0.5490
110.1265	4.65	5500	109.3663	0.5636
1.5584	5.07	6000	0.9255	0.8374
1.0883	5.49	6500	0.5841	0.8758
0.7024	5.92	7000	0.5055	0.8854
0.9033	6.34	7500	0.4639	0.8901
0.6901	6.76	8000	0.4360	0.8947
0.6114	7.19	8500	0.4080	0.8978
0.5102	7.61	9000	0.3911	0.9009
0.7154	8.03	9500	0.3747	0.9027
0.5621	8.45	10000	0.3622	0.9021
0.5262	8.88	10500	0.3554	0.9041
0.5442	9.3	11000	0.3462	0.9053
0.5615	9.72	11500	0.3416	0.9061

Framework versions

• Transformers 4.26.1
• Pytorch 1.13.1+cu117
• Datasets 2.8.0
• Tokenizers 0.13.2