bhack

Results 1415 comments of bhack

I think the only opportunity is to pass the state: https://github.com/huggingface/transformers/pull/32394#issuecomment-2276208847

If you are interested in video check https://github.com/facebookresearch/sam2/issues/3#issuecomment-2326258260

Can we add something interesting here to stress test: https://github.com/pytorch/pytorch/issues/121072 https://github.com/pytorch/pytorch/pull/121324#issuecomment-1998934523 /cc @ezyang @albanD @vfdev-5 @nWEIdia @xuzhao9 @eqy @ptrblck

Are you talking about (https://arxiv.org/pdf/1708.04552v2.pdf): ![immagine](https://user-images.githubusercontent.com/1710528/164320214-b72a5a87-ab0c-469f-958a-b4ffc4781f00.png) or something else? Edit: Ok sorry it was in the title but not in the description. It is the "fill_mode"

We have some experimental results with patch gaussian as an alternative "fill_mode" for cutout: https://arxiv.org/abs/1906.02611

The patch gaussian has a good citations threshod (>100). There is a TF implementation at page 15. I don't know if It Is better to spent time on something that...

https://github.com/tensorflow/models/blob/master/research/object_detection/core/preprocessor.py#L2793

Are we targeting 2 or 2.1?

@Hprairie I am trying to export/compile the `selective_scan` at https://github.com/pytorch/pytorch/issues/130150#issuecomment-2543153878 Do you have an exportable/compilable version?

Lingua repo hasn't selective_scsn custom_op: https://github.com/facebookresearch/lingua/tree/main/apps%2Fmamba%2Fcomponent