USS-reward-model-binary-2vs4

This model is a fine-tuned version of answerdotai/ModernBERT-large on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3419
• Accuracy: 0.9104
• F1: 0.9516
• Auc Roc: 0.9121
• Mcc: 0.4764
• Confusion Matrix: [[2, 6], [0, 59]]

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: 2
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 10
• total_train_batch_size: 20
• 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: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Auc Roc	Mcc	Confusion Matrix
4.9132	1.0	24	0.4302	0.8806	0.9365	0.7479	0.0	[[0, 8], [0, 59]]
4.2935	2.0	48	0.4398	0.8806	0.9365	0.7839	0.0	[[0, 8], [0, 59]]
3.8073	3.0	72	0.2622	0.8955	0.9421	0.8729	0.4209	[[3, 5], [2, 57]]
2.1451	4.0	96	0.3419	0.9104	0.9516	0.9121	0.4764	[[2, 6], [0, 59]]
0.9600	5.0	120	0.9652	0.8209	0.8889	0.875	0.5037	[[7, 1], [11, 48]]
0.5732	6.0	144	0.8316	0.8657	0.9244	0.8443	0.3257	[[3, 5], [4, 55]]
0.1811	7.0	168	0.9996	0.9104	0.9516	0.8559	0.4764	[[2, 6], [0, 59]]
0.2513	8.0	192	0.7501	0.8955	0.9421	0.8729	0.4209	[[3, 5], [2, 57]]
0.0001	9.0	216	0.8585	0.8955	0.9421	0.8792	0.4209	[[3, 5], [2, 57]]
0.0000	10.0	240	0.8530	0.8806	0.9333	0.8771	0.3681	[[3, 5], [3, 56]]

Framework versions

• Transformers 5.9.0
• Pytorch 2.12.0+cu130
• Datasets 4.8.5
• Tokenizers 0.22.2