Sensor AI provides decentralized GPU resources (via Golem/iExec) to train AI models using your datasets. Pay with $SENSE tokens and avoid expensive cloud costs.
Step-by-Step Guide
Step 1: Upload Your Dataset
Use your own annotated dataset or buy one from the marketplace. Ensure data is in a supported format (COCO, YOLO, etc.). Step 2: Choose a Model Framework
Sensor AI supports:
YOLOv8 (for object detection). Mask R-CNN (for instance segmentation). ResNet (for classification). Custom PyTorch/TensorFlow scripts (upload your own). Step 3: Configure Training Parameters
Epochs: How long to train (e.g., 50 epochs). Batch Size: Depends on GPU memory (e.g., 16). Learning Rate: Adjust for model convergence. Augmentations: Enable flip/rotate for better generalization. Step 4: Launch Training
Select a decentralized compute provider (e.g., Golem). Pay with $SENSE tokens (cost depends on GPU hours). Monitor progress in the "Training Jobs" dashboard. Step 5: Evaluate & Deploy
Download model weights (e.g., .pt or .h5). Test performance using validation metrics (mAP, accuracy). Sensor AI’s API (for cloud inference). Edge devices (export to ONNX/TFLite). Pro Tips
⚡ Start with a small subset to test hyperparameters.
📉 Use early stopping to save costs.
🤖 Fine-tune pre-trained models for faster results.
