vit-large-patch16-224-in21k-finetuned-galaxy10-decals

This model is a fine-tuned version of google/vit-large-patch16-224-in21k on the matthieulel/galaxy10_decals dataset. It achieves the following results on the evaluation set:

  • Loss: 0.6167
  • Accuracy: 0.8455
  • Precision: 0.8422
  • Recall: 0.8455
  • F1: 0.8432

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: 0.0001
  • train_batch_size: 128
  • eval_batch_size: 128
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 512
  • 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

Training results

Training Loss Epoch Step Validation Loss Accuracy Precision Recall F1
1.1548 0.99 31 0.8377 0.7238 0.7320 0.7238 0.6999
0.7125 1.98 62 0.6551 0.7847 0.7914 0.7847 0.7807
0.6646 2.98 93 0.5389 0.8106 0.8079 0.8106 0.8024
0.5941 4.0 125 0.5209 0.8247 0.8319 0.8247 0.8134
0.5485 4.99 156 0.5222 0.8253 0.8317 0.8253 0.8249
0.4856 5.98 187 0.5064 0.8298 0.8324 0.8298 0.8241
0.4613 6.98 218 0.4739 0.8439 0.8464 0.8439 0.8418
0.4395 8.0 250 0.4993 0.8348 0.8449 0.8348 0.8359
0.4184 8.99 281 0.5139 0.8388 0.8386 0.8388 0.8339
0.3635 9.98 312 0.5031 0.8348 0.8372 0.8348 0.8349
0.3665 10.98 343 0.5235 0.8337 0.8330 0.8337 0.8319
0.3639 12.0 375 0.5032 0.8433 0.8471 0.8433 0.8429
0.344 12.99 406 0.5632 0.8326 0.8366 0.8326 0.8295
0.3089 13.98 437 0.5355 0.8433 0.8459 0.8433 0.8415
0.2947 14.98 468 0.5596 0.8371 0.8379 0.8371 0.8363
0.279 16.0 500 0.5364 0.8433 0.8403 0.8433 0.8396
0.2465 16.99 531 0.5636 0.8377 0.8360 0.8377 0.8359
0.2404 17.98 562 0.6250 0.8343 0.8391 0.8343 0.8334
0.2365 18.98 593 0.5921 0.8410 0.8416 0.8410 0.8405
0.2307 20.0 625 0.6118 0.8393 0.8383 0.8393 0.8378
0.2217 20.99 656 0.6195 0.8410 0.8399 0.8410 0.8389
0.2104 21.98 687 0.6108 0.8326 0.8330 0.8326 0.8315
0.2079 22.98 718 0.6308 0.8405 0.8375 0.8405 0.8373
0.1784 24.0 750 0.6167 0.8455 0.8422 0.8455 0.8432
0.1901 24.99 781 0.6394 0.8439 0.8411 0.8439 0.8416
0.1801 25.98 812 0.6667 0.8348 0.8349 0.8348 0.8342
0.1892 26.98 843 0.6335 0.8393 0.8410 0.8393 0.8394
0.1756 28.0 875 0.6431 0.8455 0.8439 0.8455 0.8440
0.1791 28.99 906 0.6445 0.8444 0.8423 0.8444 0.8428
0.1728 29.76 930 0.6451 0.8439 0.8414 0.8439 0.8418

Framework versions

  • Transformers 4.37.2
  • Pytorch 2.3.0
  • Datasets 2.19.1
  • Tokenizers 0.15.1
Downloads last month
52
Safetensors
Model size
303M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for matthieulel/vit-large-patch16-224-in21k-finetuned-galaxy10-decals

Finetuned
(32)
this model