bert-for-patents-finetuned_r-as

This model is a fine-tuned version of anferico/bert-for-patents on a unique dataset consisting of 46673 patent applications, of which 11387 were defined as "RA-SYS-related" and 35000 as "Not RA-SYS-related". The fine-tuning is performed on patents' titles and abstracts. The base model was fine-tuned to perform a binary classification task, identifying patents related to the "Robotics and Autonomous Systems" domain.

It achieves the following results on the evaluation set:

• Loss: 0.0669
• Accuracy: 0.975
• Auc: 0.996
• F1: 0.947
• Precision: 0.96
• Recall: 0.935

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: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Auc	F1	Precision	Recall
0.3652	1.0	1015	0.1716	0.93	0.988	0.838	0.971	0.738
0.1382	2.0	2030	0.1070	0.959	0.993	0.912	0.97	0.861
0.1048	3.0	3045	0.0853	0.968	0.994	0.934	0.96	0.909
0.0946	4.0	4060	0.0770	0.971	0.995	0.94	0.951	0.929
0.0899	5.0	5075	0.0731	0.973	0.995	0.944	0.952	0.936
0.0854	6.0	6090	0.0706	0.972	0.996	0.943	0.958	0.928
0.0836	7.0	7105	0.0693	0.974	0.996	0.945	0.962	0.929
0.0828	8.0	8120	0.0684	0.975	0.996	0.947	0.967	0.929
0.0821	9.0	9135	0.0677	0.975	0.996	0.948	0.967	0.93
0.0794	10.0	10150	0.0669	0.975	0.996	0.947	0.96	0.935

Framework versions

• Transformers 4.52.4
• Pytorch 2.6.0+cu124
• Datasets 3.6.0.dev0
• Tokenizers 0.21.1