tokamakuV2

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

.

🧠 Model Card: Reactor Stage Classifier (Fine-tuned ResNet-50) Model Description

This model is a fine-tuned version of Microsoft’s ResNet-50 designed to classify images of a plasma reactor into three operational stages: l (low), h (high), and elm (edge-localized mode).

The dataset consists of labeled reactor images, preprocessed and augmented to improve robustness. The fine-tuning was performed using the Hugging Face transformers and datasets libraries on Google Colab, with integration to the Hugging Face Hub for model saving and version control.

🧩 Training Details

Base model: microsoft/resnet-50

Frameworks: transformers, datasets, torchvision, evaluate

Dataset format: ImageFolder (ZIP file from Google Drive, already labeled)

Dataset split: 85% training / 15% testing

Training samples: ~850 (train) / ~150 (test)

Training parameters:

Learning rate: 2e-4

Batch size: 16

Epochs: 10

Mixed precision (fp16): enabled

⚙️ Data Preprocessing & Augmentation

The preprocessing pipeline used several data augmentations to improve generalization, applied with torchvision.transforms:

Random resized crop (224×224)

Random horizontal flip

Color jitter (brightness, contrast, saturation, hue)

Random grayscale (10%)

Normalization with pretrained ResNet mean and std

Validation data was resized and center-cropped to the same input size.