git-base-naruto

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

• Loss: 0.0279
• Wer Score: 7.0134

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: 2e-05
• train_batch_size: 5
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 10
• 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
8.5536	2.2727	50	7.1184	52.625
6.2017	4.5455	100	5.0281	22.3527
4.1263	6.8182	150	2.9941	21.8616
2.2013	9.0909	200	1.2700	18.7321
0.7916	11.3636	250	0.3337	12.1607
0.1917	13.6364	300	0.0798	4.5357
0.0458	15.9091	350	0.0356	1.0
0.0142	18.1818	400	0.0278	7.25
0.0066	20.4545	450	0.0287	8.4196
0.0043	22.7273	500	0.0270	7.8795
0.0032	25.0	550	0.0272	7.2545
0.0027	27.2727	600	0.0273	7.0179
0.0023	29.5455	650	0.0271	7.2054
0.002	31.8182	700	0.0275	7.0580
0.0018	34.0909	750	0.0276	7.2589
0.0016	36.3636	800	0.0277	7.0312
0.0015	38.6364	850	0.0277	7.0759
0.0014	40.9091	900	0.0278	7.1071
0.0014	43.1818	950	0.0278	7.1161
0.0013	45.4545	1000	0.0279	6.9241
0.0013	47.7273	1050	0.0279	6.9911
0.0013	50.0	1100	0.0279	7.0134

Framework versions

• Transformers 4.40.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1