quanto
quanto copied to clipboard
huggingface
Add examples based on ViT
Vision Transformers should be supported out-of-the-box by quanto.
The goal of this issue is to add some examples under examples/vision.
At the very minimum, there should be a classification example, using a relatively small dataset so that people can rerun it locally without downloading gigabytes of images (imagenette maybe ?).
An image detection example would also be nice if possible, but maybe it requires too much code.
@dacorvo I can work on this one. I've been contributing to the transformers library recently, but I also want to get into the quanto library as it seems pretty fun!
This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.
![github-actions[bot] avatar](https://avatars.githubusercontent.com/in/15368?v=4 "Published by a pub.dev verified publisher"){kind=small}
This issue was closed because it has been stalled for 5 days with no activity.
This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.
Oh but I see now there are already 3 examples under vision. Maybe we want to add something different, like Whisper for speech recognition?
This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.
This issue was closed because it has been stalled for 5 days with no activity.
@dacorvo I am working on adding a ViT example using the oxford Pets dataset. Is that a good classification example?
Nice ! Looking forward to see your example. When submitting your pull-request, please do not forget to use conventional commits to make it easier to review.
Nice ! Looking forward to see your example. When submitting your pull-request, please do not forget to use conventional commits to make it easier to review.
I made a PR.
But I couldn't perform model requantize as I faced a problem with quantization_map function.
It seems the
QLayerNorm doesn't have a quantization weight type. Is this a issue with QLayerNorm?
It seems the
QLayerNormdoesn't have a quantization weight type. Is this a issue withQLayerNorm?
Yes, line 168 should be modified to support weight_qtype == None, as the line below. If you don't mind, you could add a fix commit to your pull-request.
Yes, I can do that. But do you think I should create a new PR for this or should I commit in the same PR?
Do it in the same pull-request, but in a separate commit. They are not squashed when merging so it will appear as a stand-alone change.
Do it in the same pull-request, but in a separate commit. They are not squashed when merging so it will appear as a stand-alone change.
I have made the change and committed. Is test case necessary for the change?
This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.