yujiepan/internal.swin-base-food101-int8-structured40

swin-food101-jpqd

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: 0.3497
• Accuracy: 0.9055

This model is quantized. Structured sparsity in transformer linear layers: 40%.

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: 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
2.2676	0.42	500	2.1087	0.7947
0.6823	0.84	1000	0.5127	0.8818
0.816	1.27	1500	0.3944	0.8954
0.5272	1.69	2000	0.3310	0.9050
12.263	2.11	2500	12.0040	0.9057
48.9519	2.54	3000	48.4500	0.8597
75.576	2.96	3500	75.5765	0.6951
93.7523	3.38	4000	93.3753	0.5992
103.7155	3.8	4500	103.5301	0.5622
107.7993	4.23	5000	108.0881	0.5636
109.6831	4.65	5500	109.2205	0.5844
1.8848	5.07	6000	0.9807	0.8315
1.0668	5.49	6500	0.6050	0.8740
0.7951	5.92	7000	0.5151	0.8838
0.7402	6.34	7500	0.4843	0.8906
0.7319	6.76	8000	0.4494	0.8933
0.5683	7.19	8500	0.4378	0.8953
0.496	7.61	9000	0.4115	0.8981
0.6174	8.03	9500	0.3952	0.9005
0.4921	8.45	10000	0.3765	0.9026
0.5843	8.88	10500	0.3678	0.9035
0.5485	9.3	11000	0.3576	0.9039
0.4337	9.72	11500	0.3512	0.9057

Framework versions

• Transformers 4.26.0
• Pytorch 1.13.1+cu116
• Datasets 2.8.0
• Tokenizers 0.13.2