|
gptkbp:instance_of
|
gptkb:Transformers
|
|
gptkbp:application
|
Image classification
|
|
gptkbp:architecture
|
gptkb:Transformers
|
|
gptkbp:available_in
|
gptkb:Hugging_Face_Model_Hub
|
|
gptkbp:batch_size
|
gptkb:32
|
|
gptkbp:can_be_fine_tuned
|
gptkb:Yes
|
|
gptkbp:coat_of_arms
|
gptkb:12
|
|
gptkbp:developed_by
|
gptkb:Facebook_AI_Research
|
|
gptkbp:drops
|
0.1
|
|
gptkbp:evaluates
|
Accuracy
Image Net validation set
|
|
gptkbp:focus_area
|
Self-attention
|
|
gptkbp:framework_version
|
1.8.0
|
|
gptkbp:headcount
|
gptkb:6
|
|
gptkbp:hidden_size
|
384
|
|
https://www.w3.org/2000/01/rdf-schema#label
|
Dei T-Small
|
|
gptkbp:image_input_size
|
224x224
|
|
gptkbp:initialization_method
|
Xavier initialization
|
|
gptkbp:input_normalization
|
Mean subtraction
|
|
gptkbp:input_output
|
Class probabilities
RGB images
|
|
gptkbp:intermediate_size
|
1536
|
|
gptkbp:is_a_framework_for
|
gptkb:Py_Torch
|
|
gptkbp:is_optimized_for
|
gptkb:Adam_W
|
|
gptkbp:is_taught_in
|
0.001
|
|
gptkbp:losses
|
Cross-entropy loss
|
|
gptkbp:max_sequence_length
|
gptkb:197
|
|
gptkbp:model
|
Small
|
|
gptkbp:model_checkpointing
|
gptkb:Yes
|
|
gptkbp:orbital_period
|
22 million
|
|
gptkbp:output_activation_function
|
Softmax
|
|
gptkbp:performance
|
Top-1 accuracy
|
|
gptkbp:predecessor
|
gptkb:Vi_T
|
|
gptkbp:provides_information_on
|
gptkb:Image_Net
gptkb:Yes
|
|
gptkbp:related_to
|
gptkb:Vision_Transformers
|
|
gptkbp:release_year
|
gptkb:2021
|
|
gptkbp:successor
|
gptkb:Dei_T-Base
|
|
gptkbp:supports_distributed_training
|
gptkb:Yes
|
|
gptkbp:supports_mixed_precision_training
|
gptkb:Yes
|
|
gptkbp:supports_multi_gputraining
|
gptkb:Yes
|
|
gptkbp:supports_quantization
|
gptkb:Yes
|
|
gptkbp:supports_transfer_learning
|
gptkb:Yes
|
|
gptkbp:top-1_accuracy
|
79.9%
|
|
gptkbp:top-5_accuracy
|
95.2%
|
|
gptkbp:training
|
Supervised learning
Less than 1 day
|
|
gptkbp:uses
|
distillation
|
|
gptkbp:uses_auto_augment
|
gptkb:Yes
|
|
gptkbp:uses_batch_normalization
|
No
|
|
gptkbp:uses_cut_mix
|
gptkb:Yes
|
|
gptkbp:uses_data_parallelism
|
gptkb:Yes
|
|
gptkbp:uses_early_stopping
|
gptkb:Yes
|
|
gptkbp:uses_ensemble_learning
|
No
|
|
gptkbp:uses_gradient_clipping
|
gptkb:Yes
|
|
gptkbp:uses_label_smoothing
|
gptkb:Yes
|
|
gptkbp:uses_layer_normalization
|
gptkb:Yes
|
|
gptkbp:uses_learning_rate_scheduler
|
gptkb:Yes
|
|
gptkbp:uses_mix_up
|
gptkb:Yes
|
|
gptkbp:uses_positional_encoding
|
gptkb:Yes
|
|
gptkbp:uses_random_erasing
|
gptkb:Yes
|
|
gptkbp:uses_residual_connections
|
gptkb:Yes
|
|
gptkbp:uses_self_supervised_learning
|
No
|
|
gptkbp:uses_semi_supervised_learning
|
No
|
|
gptkbp:uses_test_time_augmentation
|
gptkb:Yes
|
|
gptkbp:uses_transfer_learning
|
gptkb:Yes
|
|
gptkbp:uses_unsupervised_learning
|
No
|
|
gptkbp:uses_weight_decay
|
gptkb:Yes
|
|
gptkbp:bfsParent
|
gptkb:Dei_T
|
|
gptkbp:bfsLayer
|
5
|