sagemaker-bert-base-intent1018_2

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

• Loss: 0.5145
• Accuracy: 0.9017

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: 3e-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
• lr_scheduler_warmup_steps: 500
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	88	4.0951	0.0470
No log	2.0	176	3.7455	0.2158
No log	3.0	264	3.0505	0.4252
No log	4.0	352	2.0489	0.6303
No log	5.0	440	1.3342	0.7735
2.9556	6.0	528	0.9592	0.8162
2.9556	7.0	616	0.7623	0.8162
2.9556	8.0	704	0.6262	0.8547
2.9556	9.0	792	0.5145	0.9017
2.9556	10.0	880	0.5328	0.8846
2.9556	11.0	968	0.5137	0.8932
0.3206	12.0	1056	0.5190	0.8846
0.3206	13.0	1144	0.5158	0.8953
0.3206	14.0	1232	0.5053	0.8974
0.3206	15.0	1320	0.5140	0.8953
0.3206	16.0	1408	0.5108	0.8996
0.3206	17.0	1496	0.5282	0.8932
0.0381	18.0	1584	0.5278	0.8974
0.0381	19.0	1672	0.5224	0.8996
0.0381	20.0	1760	0.5226	0.8996

Framework versions

• Transformers 4.12.3
• Pytorch 1.9.1
• Datasets 1.15.1
• Tokenizers 0.10.3