Edit model card

deit-base-distilled-patch16-224-75-fold3

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

  • Loss: 0.2177
  • Accuracy: 0.9535

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: 32
  • eval_batch_size: 32
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 128
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 100

Training results

Training Loss Epoch Step Validation Loss Accuracy
No log 1.0 2 0.5697 0.6744
No log 2.0 4 0.6348 0.6977
No log 3.0 6 0.8030 0.6977
No log 4.0 8 0.7092 0.6977
0.5313 5.0 10 0.4644 0.8140
0.5313 6.0 12 0.4211 0.7907
0.5313 7.0 14 0.4913 0.8140
0.5313 8.0 16 0.4788 0.8372
0.5313 9.0 18 0.3717 0.7907
0.3672 10.0 20 0.3763 0.8140
0.3672 11.0 22 0.5082 0.8372
0.3672 12.0 24 0.3661 0.8140
0.3672 13.0 26 0.3693 0.8140
0.3672 14.0 28 0.3808 0.8605
0.2837 15.0 30 0.3492 0.8372
0.2837 16.0 32 0.3446 0.7907
0.2837 17.0 34 0.3909 0.8605
0.2837 18.0 36 0.4379 0.8372
0.2837 19.0 38 0.3905 0.8140
0.2268 20.0 40 0.3453 0.8140
0.2268 21.0 42 0.4145 0.8372
0.2268 22.0 44 0.3370 0.8372
0.2268 23.0 46 0.3502 0.8372
0.2268 24.0 48 0.3295 0.8372
0.1735 25.0 50 0.3118 0.8837
0.1735 26.0 52 0.3050 0.8837
0.1735 27.0 54 0.6940 0.8140
0.1735 28.0 56 0.5913 0.8372
0.1735 29.0 58 0.3190 0.8837
0.1221 30.0 60 0.4141 0.8372
0.1221 31.0 62 0.4572 0.8372
0.1221 32.0 64 0.3048 0.8837
0.1221 33.0 66 0.3139 0.9070
0.1221 34.0 68 0.3090 0.8837
0.1158 35.0 70 0.3393 0.8837
0.1158 36.0 72 0.3035 0.8605
0.1158 37.0 74 0.4730 0.8140
0.1158 38.0 76 0.3788 0.8605
0.1158 39.0 78 0.2904 0.8837
0.1075 40.0 80 0.2750 0.8837
0.1075 41.0 82 0.3328 0.8837
0.1075 42.0 84 0.2648 0.9070
0.1075 43.0 86 0.2517 0.8837
0.1075 44.0 88 0.4402 0.8837
0.0925 45.0 90 0.4076 0.8837
0.0925 46.0 92 0.2390 0.9070
0.0925 47.0 94 0.2176 0.9070
0.0925 48.0 96 0.2580 0.9302
0.0925 49.0 98 0.2049 0.9070
0.1085 50.0 100 0.2244 0.8837
0.1085 51.0 102 0.2377 0.9070
0.1085 52.0 104 0.4591 0.8372
0.1085 53.0 106 0.5054 0.8372
0.1085 54.0 108 0.2994 0.9302
0.0876 55.0 110 0.2387 0.9070
0.0876 56.0 112 0.3078 0.9070
0.0876 57.0 114 0.4470 0.8372
0.0876 58.0 116 0.3457 0.9070
0.0876 59.0 118 0.2655 0.9070
0.0823 60.0 120 0.2150 0.9070
0.0823 61.0 122 0.2116 0.9070
0.0823 62.0 124 0.2305 0.9302
0.0823 63.0 126 0.2070 0.9302
0.0823 64.0 128 0.1808 0.9070
0.0791 65.0 130 0.1669 0.9070
0.0791 66.0 132 0.1721 0.9070
0.0791 67.0 134 0.2194 0.9302
0.0791 68.0 136 0.3454 0.8837
0.0791 69.0 138 0.5415 0.8372
0.0607 70.0 140 0.4457 0.8605
0.0607 71.0 142 0.2411 0.8837
0.0607 72.0 144 0.2057 0.9070
0.0607 73.0 146 0.2200 0.9070
0.0607 74.0 148 0.2677 0.8837
0.0715 75.0 150 0.2950 0.8837
0.0715 76.0 152 0.2874 0.8837
0.0715 77.0 154 0.2236 0.9070
0.0715 78.0 156 0.2052 0.9302
0.0715 79.0 158 0.2177 0.9535
0.0644 80.0 160 0.2178 0.9535
0.0644 81.0 162 0.2126 0.9302
0.0644 82.0 164 0.2127 0.9302
0.0644 83.0 166 0.2216 0.9070
0.0644 84.0 168 0.2420 0.9070
0.0622 85.0 170 0.2305 0.9070
0.0622 86.0 172 0.2247 0.9070
0.0622 87.0 174 0.2492 0.9070
0.0622 88.0 176 0.3292 0.8837
0.0622 89.0 178 0.3876 0.8837
0.0564 90.0 180 0.3886 0.8837
0.0564 91.0 182 0.3707 0.8837
0.0564 92.0 184 0.3377 0.8837
0.0564 93.0 186 0.3186 0.8837
0.0564 94.0 188 0.3038 0.8837
0.0578 95.0 190 0.2818 0.8605
0.0578 96.0 192 0.2756 0.8837
0.0578 97.0 194 0.2694 0.8837
0.0578 98.0 196 0.2698 0.8837
0.0578 99.0 198 0.2732 0.8837
0.0424 100.0 200 0.2739 0.8837

Framework versions

  • Transformers 4.41.0
  • Pytorch 2.3.0+cu121
  • Datasets 2.19.1
  • Tokenizers 0.19.1
Downloads last month
2
Safetensors
Model size
85.8M params
Tensor type
F32
·
Inference API
Drag image file here or click to browse from your device
This model can be loaded on Inference API (serverless).

Finetuned from

Evaluation results