You have to remove the .SVG files before you start, from your dataset.
As far as I am aware, most vision models use raster based file formats, that are pixel based such as .jpg and .png. Svg files are vector based files, that is they use geometrical expressions written in XML. If you open up an .svg file you will see it looks similar to an html page it has text in it.
In relation to the fastai vision library trying to classify an .svg file, is like creating a text file with “this is an image file” inside and then trying to recognise the file as an image.
So either remove every .svg file from your dataset or convert them to jpg or png and then use them to train your model.