skin-mole-vit-onnx

ONNX export (fp32 + 8-bit quantized) of Anwarkh1/Skin_Cancer-Image_Classification, a ViT-base fine-tuned on HAM10000 (7 skin-lesion classes), packaged for transformers.js in-browser inference.

Quantized model verified to match fp32 predictions on held-out dermoscopy images (<1% probability shift). Used by the MoleCheck Space.

Not a medical device — research/education use only.

import { pipeline } from "@huggingface/transformers";
const classifier = await pipeline("image-classification", "sazio/skin-mole-vit-onnx", { dtype: "q8" });
const preds = await classifier(imageUrl, { top_k: 7 });