Eugene Zatepyakin

the current implementation is a good reference how ever it will not compile with Apple Clang for example and wont with MSVC... but if you try to fix it for...

well it irritates as much as it possible :) but unfortunately thats life developing multi-platform solutions. u have to compile using native toolchains so customers can integrate it easily in...

good point will extend hit area :)

you are probably rotating camera without any translation this kind of movement is a killer for most monocular slam systems. try to shift camera instead of rotating it. also u...

i think having prefix and query dynamically sized is the best for robustness as well as inference usage

yup, when training discriminator it looks good, i was referring generator training block, sorry for the confusion

grouped_sims never used actually :) https://github.com/lucidrains/tranception-pytorch/blob/610ebf21c4cf3b7f0d560c89610051c7242527a0/tranception_pytorch/tranception_pytorch.py#L165 also Alibi impl here has an issue with backward call complaining about missing/overwritten gradient, after replacing it with another impl from x-transformers repo it...

It seems the issue with ALiBi is due to few network forward calls before backward call, specifically i have ``` p0 = net(x) with torch.no_grad(): p = net(y) loss =...

mask needs to be moved to correct device here: https://github.com/lucidrains/tranception-pytorch/blob/a09cba2a66952ce4fb0f57f8441cba0b1eb15d37/tranception_pytorch/tranception_pytorch.py#L177 i also think its good to have it optional in case we dont need causual masking?

i noticed that you apply alibi here inside forward: https://github.com/lucidrains/tranception-pytorch/blob/128400a53e69e9691c9c593e38233d20a353aee3/tranception_pytorch/tranception_pytorch.py#L52 however the interface of function assume it will not apply bias but will return it instead for manual addition in...