[docs] ViTPose #38630
[docs] ViTPose #38630
Conversation
2bff6a5 to
f7abea2
Compare
|
Sorry for the delay @NielsRogge, but I think it should be resolved here. Just trying to figure out why the |
docs/source/en/model_doc/vitpose.md
Outdated
|
|
||
| inputs = person_image_processor(images=image, return_tensors="pt").to(device) | ||
| inputs = person_image_processor(images=image, return_tensors="pt").to("cuda") |
There was a problem hiding this comment.
We typically don't add "cuda" by default for users who want to run this on CPU
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
| <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/vitpose.png"/> | ||
| </div> | ||
|
|
||
| Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends. |
There was a problem hiding this comment.
Quantization seems indeed useful for large (language) models, but ViTPose are pretty small models you can run on-device. The weights are about 300MB. Any reason to push for quantization here?
There was a problem hiding this comment.
Yeah, I included it in case users want to run the vitpose-plus-huge checkpoint! Those weights are about 3.6GB. Let me know if this is large enough to warrant quantization, otherwise happy to remove
There was a problem hiding this comment.
Ok sounds good (wouldn't recommend it for the base or large versions)
|
Let me know if there any other changes @NielsRogge , otherwise lets merge it! 😄 |
Updates model card for ViTPose and fixes #36773