review_classification_bert_base_jp_v3_ratio1_5_2label_add_dropout-epoch15_v4

This model is a fine-tuned version of cl-tohoku/bert-base-japanese-v3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.2584
• Accuracy: 0.8235

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: 1e-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: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	156	0.3394	0.8460
No log	2.0	312	0.3416	0.8338
No log	3.0	468	0.3775	0.8338
0.2942	4.0	624	0.4964	0.8319
0.2942	5.0	780	0.6834	0.8263
0.2942	6.0	936	0.8556	0.8376
0.0772	7.0	1092	0.9792	0.8197
0.0772	8.0	1248	1.0804	0.8188
0.0772	9.0	1404	1.1236	0.8244
0.0082	10.0	1560	1.1708	0.8254
0.0082	11.0	1716	1.1788	0.8272
0.0082	12.0	1872	1.1949	0.8310
0.002	13.0	2028	1.2106	0.8282
0.002	14.0	2184	1.2539	0.8235
0.002	15.0	2340	1.2584	0.8235

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0