Llama-3.1-8B-medquad-V2

This model is a fine-tuned version of meta-llama/Llama-3.1-8B on the MedQuAD: Ben-Abacha and Demner-Fushman (2019) dataset. It achieves the following results on the evaluation set:

• Loss: 0.8959

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: 0.0002
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 12
• total_train_batch_size: 192
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: reduce_lr_on_plateau
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 7
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
1.2503	0.1462	10	1.1359
1.1182	0.2923	20	1.0199
1.0864	0.4385	30	0.9856
0.9031	0.5847	40	0.9681
1.0773	0.7308	50	0.9499
0.9575	0.8770	60	0.9427
0.9768	1.0231	70	0.9452
0.9673	1.1693	80	0.9264
0.8541	1.3155	90	0.9282
0.9772	1.4616	100	0.9180
0.8427	1.6078	110	0.9211
0.9317	1.7540	120	0.9142
0.9498	1.9001	130	0.9011
0.8412	2.0463	140	0.9036
0.899	2.1924	150	0.9031
0.7488	2.3386	160	0.8990
0.8824	2.4848	170	0.9033
0.8334	2.6309	180	0.8959

Framework versions

• PEFT 0.13.0
• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0