vit-base-patch16-224-imigue

This model is a fine-tuned google/vit-base-patch16-224-in21k, on TornikeO/imigue micro-level emotion classification dataset. The evaluation performance is as follows (per-class evals are precisions for that particular class. F1 score is micro-averaged.):

• eval_loss: 0.6450
• eval_accuracy: 0.8112
• eval_f1: 0.6905
• eval_arms_akimbo: 1.0
• eval_biting_nails: 0.0
• eval_buckle_button,_pulling_shirt_collar,_adjusting_tie: 0.8923
• eval_bulging_face,_deep_breath: 0.6162
• eval_covering_face: 0.8788
• eval_crossing_fingers: 0.8468
• eval_dustoffing_clothes: 0.77
• eval_folding_arms: 0.7598
• eval_head_up: 0.8182
• eval_hold_back_arms: 0.7015
• eval_illustrative_body_language: 0.8521
• eval_minaret_gesture: 0.9677
• eval_moving_torso: 0.7914
• eval_playing_with_or_adjusting_hair: 0.8393
• eval_playing_with_or_manipulating_objects: 0.9053
• eval_pressing_lips: 0.7363
• eval_putting_arms_behind_body: 0.0
• eval_rubbing_eyes: 0.8793
• eval_rubbing_or_holding_hands: 0.8180
• eval_scratching_back: 0.875
• eval_scratching_or_touching_arms: 0.7704
• eval_shaking_shoulders: 0.7051
• eval_sitting_upright: 0.7273
• eval_touching_ears: 0.8261
• eval_touching_hat: 0.9474
• eval_touching_jaw: 0.8979
• eval_touching_or_covering_suprasternal_notch: 1.0
• eval_touching_or_scratching_facial_parts: 0.8178
• eval_touching_or_scratching_forehead: 0.8
• eval_touching_or_scratching_head: 0.8913
• eval_touching_or_scratching_neck: 0.8788
• eval_turtle_neck: 1.0
• eval_runtime: 13.9155
• eval_samples_per_second: 869.752
• eval_steps_per_second: 3.449
• step: 0

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: 128
• eval_batch_size: 256
• 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: 20
• mixed_precision_training: Native AMP

Framework versions

• Transformers 4.39.2
• Pytorch 2.2.1
• Datasets 2.18.0
• Tokenizers 0.15.2