CodeGenDetect-CodeBert_Lora

This model is a fine-tuned version of microsoft/codebert-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0384
• Accuracy: 0.9907
• F1: 0.9907
• Precision: 0.9907
• Recall: 0.9907

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Accuracy	F1	Validation Loss	Precision	Recall
0.1381	0.128	4000	0.9586	0.9586	0.1627	0.9599	0.9586
0.0821	0.256	8000	0.9761	0.9761	0.1081	0.9761	0.9761
0.0667	0.384	12000	0.9786	0.9786	0.1008	0.9787	0.9786
0.0754	0.512	16000	0.9820	0.9820	0.0779	0.9821	0.9820
0.0776	0.64	20000	0.9846	0.9846	0.0617	0.9847	0.9846
0.0643	0.768	24000	0.9831	0.9831	0.0761	0.9832	0.9831
0.064	0.896	28000	0.9878	0.9878	0.0495	0.9878	0.9878
0.0477	1.024	32000	0.9879	0.9879	0.0480	0.9880	0.9879
0.0427	1.152	36000	0.9894	0.9894	0.0424	0.9894	0.9894
0.0381	1.28	40000	0.9880	0.9880	0.0484	0.9880	0.9880
0.0423	1.408	44000	0.9901	0.9901	0.0399	0.9901	0.9901
0.0389	1.536	48000	0.9888	0.9888	0.0513	0.9889	0.9888
0.0416	1.6640	52000	0.9908	0.9908	0.0358	0.9908	0.9908
0.0374	1.792	56000	0.0370	0.9905	0.9905	0.9905	0.9905
0.0441	1.92	60000	0.0355	0.9905	0.9905	0.9905	0.9905
0.0358	2.048	64000	0.0384	0.9907	0.9907	0.9907	0.9907

Framework versions

• PEFT 0.18.0
• Transformers 4.57.3
• Pytorch 2.9.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.1