git-base-one-5e-5-25

This model is a fine-tuned version of microsoft/git-base on the imagefolder dataset. It achieves the following results on the evaluation set:

• Loss: 3.8238
• Wer Score: 5.8

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Score
No log	1.25	5	9.5084	50.1
9.9196	2.5	10	8.4842	64.85
9.9196	3.75	15	7.9215	67.6
8.0697	5.0	20	7.4826	66.45
8.0697	6.25	25	7.0776	53.95
7.2067	7.5	30	6.6926	18.05
7.2067	8.75	35	6.3268	17.6
6.4594	10.0	40	5.9807	19.8
6.4594	11.25	45	5.6568	19.35
5.7908	12.5	50	5.3563	6.15
5.7908	13.75	55	5.0803	6.2
5.2135	15.0	60	4.8305	5.8
5.2135	16.25	65	4.6068	5.75
4.7358	17.5	70	4.4111	5.8
4.7358	18.75	75	4.2427	5.8
4.3652	20.0	80	4.1027	5.8
4.3652	21.25	85	3.9908	5.8
4.1076	22.5	90	3.9070	5.8
4.1076	23.75	95	3.8515	5.8
3.9616	25.0	100	3.8238	5.8

Framework versions

• Transformers 4.44.2
• Pytorch 2.5.0+cu121
• Datasets 3.0.2
• Tokenizers 0.19.1