Modelpublic

nclgbd/medsiglip-448-pneumonia-finetune

sourceHugging Faceotherupdated 1y agoView on Hugging Face

2likes40downloads

Model Card

medsiglip-448

This model is a fine-tuned version of google/medsiglip-448 on the vllm-pneumonia-detection/mimic-cxr-labelled-dataset dataset. It achieves the following results on the evaluation set:

—Loss: 0.2236
—F1-score: 0.7601
—Sensitivity: 0.4673
—Specificity: 0.9754
—Recall: 0.7213
—Precision: 0.8271
—Accuracy: 0.9135

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: 5e-06
—trainbatchsize: 96
—evalbatchsize: 96
—seed: 42
—optimizer: Use OptimizerNames.ADAMWTORCHFUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
—lrschedulertype: cosine
—num_epochs: 25

Training results

Training Loss	Epoch	Step	Validation Loss	F1-score	Sensitivity	Specificity	Recall	Precision	Accuracy
0.2792	0.4363	500	0.2565	0.733	0.4112	0.9754	0.6933	0.8106	0.9067
0.2243	0.8726	1000	0.2367	0.7376	0.4206	0.9754	0.698	0.8135	0.9078
0.2088	1.3089	1500	0.2167	0.7792	0.5047	0.9767	0.7407	0.8422	0.9192
0.2079	1.7452	2000	0.2220	0.7525	0.4112	0.9883	0.6998	0.877	0.9181
0.1989	2.1815	2500	0.2236	0.7601	0.4673	0.9754	0.7213	0.8271	0.9135

Framework versions

—Transformers 4.54.0
—Pytorch 2.8.0+cu128
—Datasets 3.6.0
—Tokenizers 0.21.4