Rui Wang

Hi, We have been looking into this issue and found one problem with RoPE not necessarily regarding this problem, namely that it is symmetric, which may not be what we...

Hi, Sorry to bother you but if I have trained a Unigram tokenizer using its `UnigramTrainer`, am I correct to understand that the values following the token in the json...

Hopefully yes! Stay tuned!

Hi, Stay tuned we have something cooking. Rui Wang Helixon Research On Oct 30, 2023 at 9:46 AM +0800, Johnny Tam ***@***.***>, wrote: Hello, any updates on this : )...

Hi, Sorry we do not have access to A40 GPUs so we do not think we could answer this.

Hi, I'd like to say this is convenient to us when we try to refactor projects that look like [OpenFold](https://github.com/aqlaboratory/openfold/tree/6908936b68ae89f67755240e2f588c09ec31d4c8), where we may have different number of batched dimensions. For...

Hi, 1. You could do that but Shapley value computation is extremely hard if not decomposed into several feature subgroups. 2. It does not seem to be possible, at least...

Hi, We have something in the works but not a time target yet. Stay tuned!

Hi, Sorry we do not yet support this. Will look into it!

Hi, During training we are inputting the cropped sequence. As for the ablations @mooninrain