Phil Wang

@Baran-phys hey Baran, thanks for sharing your paper. it is interesting but i will probably not accept as it is not relevant for this repository. periodic activation functions is something...

@TomKoester does it even work for single class predictions? happy to build it if you can show me your results

hey Garrett at Madison! beautiful city, still have fond memories of it (worked at Epic Systems for a year right out of college) yup, i think i may have an...

@GarrettMerz basically i'm incorrectly caching by the sequence length, but it should cache the longest sequence length and slice out any subsequent calls with shorter ones

@GarrettMerz want to give 0.5.0 a try and see if it still hangs?

@GarrettMerz sounds good, as long as it does not hang anymore best with your research and life out in the midwest

hmm, yea, i'll wait for more info from your end you are the only one reporting this

@GarrettMerz could you try turning off cache altogether? https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py#L82 just to confirm that it is indeed caused by the freqs caching and not something on your end?

@lunixbochs I see! thank you for this info I'll try a way of standardizing the cache to same tensor shape across devices and ping you to give it a try...

@lunixbochs sounds good, I'll take a look only if I can't figure it out