lucidrains / hourglass-transformer-pytorch Goto Github PK

@lucidrains Just Awesome to see this super fast launch !

The third example under "For Images...." throws the following error .,

Then , when I gave the inputs with num_tokens = 20000,
max_seq_len = 1024, I got the following runtime error

RuntimeError: Given normalized_shape=[512], expected input with shape [*, 512], but got input of size...

it keeps saying that it expected input with shape [*, 512] but i got a 4dim or 3dim input
tried flattening but still some dimension issues

i'm trying to see the benchmarks for image classification, would be grateful for any help

For the ImageNet64 generation experiments, the original authors used rotary positional embeddings. I think it would be worth adding this feature to the repository.

I can prepare a pull request that solves this~