Joshua Lochner

Thanks so much! This is amazing! I think it would be good to also name/suffix these models slightly differently, so you can export (and use) multiple versions. Also, to help...

@bil-ash ORT recently added support for `MatMulNBit` and will be available in v1.17.3. You can already access it with [this version](https://www.npmjs.com/package/onnxruntime-web/v/1.18.0-dev.20240309-efad5bbc5a).

> Did you mean the model file name like model_quantized_int4.onnx or model_quantized_nf4.onnx ? Yes, exactly :) I think this PR can help standardize which suffixes we should use (as `quantized`...

Thanks for this contribution 🤗! Since 4-bit quantization will only be supported in v3, I've merged it into [that branch](https://github.com/xenova/transformers.js/pull/545/commits/9efbada2b97fdb22fb0775ebbe30d93371cb83a3).

Here's a tiny randomly-initialized StableLM text-classification model: https://huggingface.co/hf-internal-testing/tiny-random-StableLmForSequenceClassification which you can use to test with!

Hi! 👋 1. The [Hugging Face Hub](https://huggingface.co/models) is the go-to place for sharing your machine learning projects: models, datasets, apps, you name it! There are currently over [240k open-source models](https://huggingface.co/models?sort=downloads)...

Can you try surrounding the arguments with double quotes? e.g., ```sh chat_downloader "https://www.youtube.com/watch?v=-ZLltnPt5WM" -o "-ZLltnPt5WM.json" -q ```

In that case, can you try: 1. appending a space to the beginning of the video id: `' -ZLltnPt5WM.json'` 2. escape the dash (I'm not sure how, perhaps `\`?

Just to confirm, are the model weights licensed as GPL3 or AGPL3? As stated on https://www.ultralytics.com/license (FAQ) for YOLOv8: > **Are Ultralytics YOLO-trained models licensed under the AGPL- 3.0 license?**...

Still an issue with shapes other than what the model is exported with: ``` $ optimum-cli export onnx --model hustvl/vitmatte-small-distinctions-646 o --task image-matting Framework not specified. Using pt to export...