longformer-base-4096-bible

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0001
• Macro Average Precision: 1.0
• Weighted Average Precision: 1.0000

Model description

Fine-Tuned allenai/longformer-base-4096 model to predict Bible chapter relevance from 1 to 5.

Intended uses & limitations

Predict relevance from 1 to 5 for Bible text. Fine-Tune for each query separately.

Training and evaluation data

One query. Chapter ID. Chapter Name. Chapter Text. Relevance Label.

Training procedure

Train for 12 epochs adjusting learning rate by cosine with 2 restarts and 4 epochs of warmup.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 8
• eval_batch_size: 8
• 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: cosine_with_restarts
• lr_scheduler_warmup_steps: 60
• num_epochs: 12

Training results

Training Loss	Epoch	Step	Validation Loss	Macro Average Precision	Weighted Average Precision
0.0536	1.0	58	0.0210	0.6129	0.8579
0.0339	2.0	116	0.0113	0.8049	0.9465
0.0147	3.0	174	0.0047	0.9481	0.9867
0.0065	4.0	232	0.0014	1.0	1.0000
0.0024	5.0	290	0.0008	1.0	1.0000
0.0014	6.0	348	0.0006	1.0	1.0000
0.0013	7.0	406	0.0006	1.0	1.0000
0.002	8.0	464	0.0004	1.0	1.0000
0.0007	9.0	522	0.0003	1.0	1.0000
0.0003	10.0	580	0.0001	1.0	1.0000
0.0002	11.0	638	0.0001	1.0	1.0000
0.0002	12.0	696	0.0001	1.0	1.0000

Framework versions

• Transformers 4.50.3
• Pytorch 2.6.0+cu124
• Datasets 2.18.0
• Tokenizers 0.21.1