USS-reward-model-baseline_l1

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: 0.1565
• Mse: 0.1842
• Mae: 0.3174
• R2: 0.0451
• Spearman Correlation: 0.2187

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	Mse	Mae	R2	Spearman Correlation
3.3516	1.0	97	0.1584	0.1863	0.3310	0.0343	nan
1.8064	2.0	194	0.2139	0.2374	0.4358	-0.2306	0.1389
1.4958	3.0	291	0.1842	0.2079	0.3739	-0.0777	0.1741
1.1016	4.0	388	0.1668	0.1920	0.3385	0.0048	0.2140
0.7270	5.0	485	0.1574	0.1897	0.3120	0.0163	0.1857
0.4099	6.0	582	0.1947	0.2221	0.3877	-0.1517	0.2582
0.2262	7.0	679	0.1598	0.1982	0.3105	-0.0275	0.2665
0.1335	8.0	776	0.1585	0.1877	0.3180	0.0271	0.1926
0.0647	9.0	873	0.1563	0.1843	0.3170	0.0444	0.1971
0.0282	10.0	970	0.1565	0.1842	0.3174	0.0451	0.2187

Framework versions

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