bert-fraud-classification-test-mass

This model is a fine-tuned version of google-bert/bert-base-multilingual-uncased on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3963
• F1: 0.8194
• Precision: 0.8445
• Val Accuracy: 0.8375

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: 44
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 88
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	F1	Precision	Val Accuracy
0.5197	0.1743	40	0.5468	0.7488	0.6907	0.7459
0.5208	0.3486	80	0.4667	0.7687	0.7890	0.7911
0.4235	0.5229	120	0.4351	0.7986	0.7898	0.8113
0.404	0.6972	160	0.4577	0.7972	0.7751	0.8066
0.3736	0.8715	200	0.4274	0.7914	0.8775	0.8240
0.419	1.0458	240	0.4058	0.7912	0.8737	0.8232
0.2701	1.2200	280	0.4075	0.8124	0.8393	0.8316
0.4345	1.3943	320	0.4246	0.8110	0.8088	0.8244
0.3258	1.5686	360	0.4023	0.7992	0.8788	0.8294
0.3938	1.7429	400	0.3945	0.8174	0.8447	0.8361
0.2529	1.9172	440	0.3963	0.8194	0.8445	0.8375

Framework versions

• Transformers 4.46.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0