TensorflowJS Load Model inside stackblitz ?

[nukadelic]

Since static assets are not yet supported over here, I have converted my weights file ( "group1-shard1of1", no file extension ) to base64 and saved it into a json file. From there it was possible to load the file via import as a string. Don't know much about binary arrays, but i did try base64-js lib, the fetch function and converting the data to blobs. Nothing seems to work for so far...

Is there a way to convert base64 into browser uploaded file objects ? Or any other solution ?

I'd really love to run my experiments in your editor, please consider adding support for / files as well, the one i'm trying to load is only 1/Mb in size.

Reference:

• Medium Article
• My attempt so far: https://tf-mnist.stackblitz.io ( original: source code)
• Loading tf model
• Base64 to file
• Keras HDF5 file format
• TF Weights loader, Http test
• [Encode / Decode via base64-js](https://developer.mozilla.org/en-US/docs/Web/API/WindowBase64/Base64_encoding_and_decoding#Solution_2_%E2%80%93_rewrite_the_DOMs_atob()_and_btoa()_using_JavaScript's_TypedArrays_and_UTF-8)
• Problems converting blob to JSON
• Fetch from base64 ...
• Online tool to convert any file to Base64 string

---

Edit: See working solution on codesandbox

[kc0tlh]

It looks like it's almost working in this demo app, but not quite there: https://stackblitz.com/edit/tensorflowjs-demo