git-base-CocoCaptions

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

• Loss: 0.2839
• Wer Score: 1.0

Model description

More information needed

Intended uses & limitations

University researcg project

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.01
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Score
No log	0.8	100	0.2959	1.0
No log	1.6	200	0.2859	1.0
No log	2.4	300	0.2821	1.0
No log	3.2	400	0.2863	1.0
0.4331	4.0	500	0.2833	1.0
0.4331	4.8	600	0.2837	1.0
0.4331	5.6	700	0.2827	1.0
0.4331	6.4	800	0.2850	1.0
0.4331	7.2	900	0.2870	1.0
0.2566	8.0	1000	0.2842	1.0
0.2566	8.8	1100	0.2841	1.0
0.2566	9.6	1200	0.2839	1.0

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.1+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0