StanfordAIMI/GREEN-RadLlama2-7b

StanfordAIMI/GREEN

This model is a fine-tuned version of StanfordAIMI/RadLLaMA-7b. It achieves the following results on the evaluation set:

• Loss: 0.0644

Model description and Training procedure

Please see the project website at https://stanford-aimi.github.io/green.html.

Intended uses & limitations

This model is finetuned to evaluate the difference between the reference and candidate radiology report for Chest Xrays.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 32
• total_train_batch_size: 2048
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.2634	0.64	25	0.2924
0.1216	1.28	50	0.0898
0.0833	1.92	75	0.0718
0.062	2.56	100	0.0644

Framework versions

• Transformers 4.38.0.dev0
• Pytorch 2.2.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1