Llama-3.1-8B-medquad-V3

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: 0.9213

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: 0.0001
• train_batch_size: 20
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 12
• total_train_batch_size: 240
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 7
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
1.6067	0.1826	10	1.5411
1.5616	0.3653	20	1.2822
1.2619	0.5479	30	1.1497
1.1906	0.7306	40	1.0566
1.0764	0.9132	50	0.9903
0.9496	1.0959	60	0.9758
1.0131	1.2785	70	0.9630
0.9908	1.4612	80	0.9502
0.9786	1.6438	90	0.9434
0.9182	1.8265	100	0.9366
0.9621	2.0091	110	0.9341
0.9724	2.1918	120	0.9254
0.8955	2.3744	130	0.9213

Framework versions

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