Keyurjotaniya007/microsoft-swin-tiny-img-emotion

swin-tiny

This model is a fine-tuned version of microsoft/swin-tiny-patch4-window7-224 on a Keyurjotaniya007/img-emotion-classification dataset. It achieves the following results on the evaluation set:

• Eval loss: 1.108536958694458

• Eval Accuracy: 0.5860458533823946

Training hyperparameters

The following hyperparameters were used during training:

• remove_unused_columns=False

• learning_rate=3e-6

• per_device_train_batch_size=64

• gradient_accumulation_steps=4

• per_device_eval_batch_size=64

• num_train_epochs=20

• warmup_steps=50

• weight_decay=0.02

• metric_for_best_model="accuracy"

• fp16=True

• lr_scheduler_type= "cosine"