git-base-pokemon

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: 0.0407
• Wer Score: 2.1746

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Score
7.3192	2.13	50	4.4908	21.3506
2.2486	4.26	100	0.3526	10.3803
0.1047	6.38	150	0.0321	0.3635
0.0222	8.51	200	0.0298	0.6636
0.0145	10.64	250	0.0286	2.1759
0.0081	12.77	300	0.0321	2.9690
0.004	14.89	350	0.0340	2.2962
0.002	17.02	400	0.0356	2.1837
0.0012	19.15	450	0.0370	3.1501
0.0009	21.28	500	0.0379	2.5821
0.0007	23.4	550	0.0382	2.7995
0.0006	25.53	600	0.0386	2.8318
0.0006	27.66	650	0.0387	2.4541
0.0006	29.79	700	0.0390	2.6404
0.0006	31.91	750	0.0395	2.5614
0.0006	34.04	800	0.0395	2.5317
0.0006	36.17	850	0.0399	2.5886
0.0006	38.3	900	0.0403	2.3829
0.0006	40.43	950	0.0404	2.2937
0.0006	42.55	1000	0.0404	2.2173
0.0006	44.68	1050	0.0406	2.1617
0.0006	46.81	1100	0.0406	2.1669
0.0006	48.94	1150	0.0407	2.1746

Framework versions

• Transformers 4.27.4
• Pytorch 1.13.1
• Datasets 2.11.0
• Tokenizers 0.13.3