metadata
base_model:
- apple/mobilevit-small
model_name: mobilevit_small-brain_tumor
tags:
- sft
- vision
- medical-imaging
- brain-tumor
license: apache-2.0
language:
- en
pipeline_tag: image-classification
library_name: transformers
🧠MobileViT-Small — Brain Tumor Classifier (LoRA Fine-tune)
This model is a LoRA fine-tuned version of apple/mobilevit-small
for brain tumor classification on MRI images from the BRISC2025 dataset.
🧩 Configuration
| Attribute | Value |
|---|---|
| Base Model | apple/mobilevit-small |
| Fine-tuning Method | LoRA (Low-Rank Adaptation) |
| Dataset | BRISC2025 |
| Classes | Glioma, Meningioma, No Tumor, Pituitary |
| Epochs | 10 |
| Batch Size | 32 |
| Learning Rate | 0.0005 |
| Optimizer | AdamW |
| LoRA Config | r=4, α=32, dropout=0.1, target_modules=[query, value] |
| Trainable Params | 30.9K / 4.97M (0.62%) |
🧠Example Image
📊 Evaluation Results
| Metric | Avg | Glioma | Meningioma | No Tumor | Pituitary |
|---|---|---|---|---|---|
| Accuracy | 0.9797 | 0.9750 | 0.9578 | 0.9969 | 0.9797 |
| Precision | 0.9570 | 0.9708 | 0.9419 | 0.9839 | 0.9316 |
| Recall | 0.9570 | 0.9379 | 0.8902 | 1.0000 | 1.0000 |
| F1 Score | 0.9565 | 0.9540 | 0.9154 | 0.9919 | 0.9646 |
| AUC | 0.9980 | 0.9980 | 0.9943 | 0.9999 | 0.9999 |
Test loss: 0.1146  Inference time: 0.436 s