deberta-toxic-reward-grpo2

This model is a fine-tuned version of microsoft/deberta-v3-base on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3737
• Accuracy: 0.8049
• Precision: 0.7432
• Recall: 0.9948
• F1: 0.8508
• Auc: 0.8926

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: 64
• eval_batch_size: 16
• seed: 13
• gradient_accumulation_steps: 8
• total_train_batch_size: 512
• 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
• num_epochs: 8

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Precision	Recall	F1	Auc
No log	1.0	71	0.3875	0.7164	0.8763	0.5741	0.6937	0.8758
No log	2.0	142	0.3601	0.8029	0.7437	0.9881	0.8487	0.8898
No log	3.0	213	0.3846	0.8026	0.7449	0.9843	0.8480	0.8906
No log	4.0	284	0.3749	0.8035	0.7458	0.9843	0.8486	0.8912
No log	5.0	355	0.3739	0.8050	0.7434	0.9948	0.8509	0.8927
No log	6.0	426	0.3768	0.8055	0.7452	0.9913	0.8508	0.8938
No log	7.0	497	0.3784	0.8064	0.7673	0.9386	0.8444	0.8941
0.3565	8.0	568	0.3834	0.8071	0.7653	0.9450	0.8457	0.8942

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.1+cu130
• Datasets 4.4.1
• Tokenizers 0.22.1