convnext-base-224_finetuned_on_unlabelled_IA_with_snorkel_labels

This model is a fine-tuned version of facebook/convnext-base-224 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3443
• Precision: 0.9864
• Recall: 0.9822
• F1: 0.9843
• Accuracy: 0.9884

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10
• mixed_precision_training: Native AMP
• label_smoothing_factor: 0.2

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.3611	1.0	2021	0.3467	0.9843	0.9729	0.9784	0.9842
0.3524	2.0	4042	0.3453	0.9853	0.9790	0.9821	0.9868
0.3466	3.0	6063	0.3438	0.9854	0.9847	0.9851	0.9889
0.3433	4.0	8084	0.3434	0.9850	0.9808	0.9829	0.9873
0.3404	5.0	10105	0.3459	0.9853	0.9790	0.9821	0.9868
0.3384	6.0	12126	0.3453	0.9853	0.9790	0.9821	0.9868
0.3382	7.0	14147	0.3437	0.9864	0.9822	0.9843	0.9884
0.3358	8.0	16168	0.3441	0.9857	0.9829	0.9843	0.9884
0.3349	9.0	18189	0.3448	0.9857	0.9829	0.9843	0.9884
0.3325	10.0	20210	0.3443	0.9864	0.9822	0.9843	0.9884

Framework versions

• Transformers 4.22.2
• Pytorch 1.12.1+cu113
• Datasets 2.5.1
• Tokenizers 0.12.1