medsiglip-appendicitis-binary

This model is a fine-tuned version of google/medsiglip-448 on the imagefolder dataset. It achieves the following results on the evaluation set:

• Loss: 2.4233
• Accuracy: 0.8143
• F1: 0.8056
• Roc Auc: 0.8840
• Sensitivity: 0.8007
• Specificity: 0.8225

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use adamw_bnb_8bit with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Accuracy	F1	Validation Loss	Roc Auc	Sensitivity	Specificity
2.1622	0.5253	500	0.7215	0.7215	0.6635	0.8816	0.9502	0.5836
1.2676	1.0504	1000	0.8161	0.8071	0.6515	0.9006	0.7976	0.8273
1.1505	1.5758	1500	0.8164	0.8026	0.7922	0.8917	0.7345	0.8658
0.6511	2.1009	2000	0.8130	0.7991	1.0340	0.8875	0.7314	0.8622
0.8176	2.6262	2500	0.7980	0.7809	1.2022	0.8752	0.6889	0.8639
0.6471	3.1513	3000	0.8291	0.8159	1.6653	0.8943	0.7463	0.8790
0.3691	3.6766	3500	0.8076	0.8007	1.6310	0.8846	0.8268	0.7960
0.4232	4.2017	4000	0.8079	0.7995	1.8503	0.8832	0.8021	0.8113
0.5992	4.7271	4500	0.8151	0.7937	2.6899	0.8863	0.6557	0.9111
0.2965	5.2522	5000	0.8225	0.8084	2.4061	0.8953	0.7317	0.8773
0.1884	5.7775	5500	0.8157	0.7996	2.4848	0.8841	0.7070	0.8813
0.1886	6.3026	6000	0.8083	0.7906	2.7710	0.8766	0.6889	0.8803
0.1329	6.8279	6500	2.8092	0.7984	0.7733	0.8743	0.6192	0.9065
0.1503	7.3530	7000	3.1173	0.8021	0.7780	0.8720	0.6286	0.9067
0.1636	7.8784	7500	2.9567	0.7937	0.7653	0.8740	0.5923	0.9151
0.2007	8.4035	8000	2.3409	0.8210	0.8082	0.8828	0.7481	0.8649
0.2111	8.9288	8500	2.4233	0.8143	0.8056	0.8840	0.8007	0.8225

Framework versions

• Transformers 5.15.0
• Pytorch 2.13.0+cu130
• Datasets 5.0.1
• Tokenizers 0.22.2