Dmitrii Magas

Sorry for late response, @matthew-brett All packages were installed via pip3, ubuntu 16.04.2, python 3.5.2

+1, and it's already logged on the cpp server side, just has to be propagated into the API response: https://github.com/ollama/ollama/blob/1963c00201958da7165a40f9d2f22b28e11be718/llm/ext_server/server.cpp#L1682

Same error, can be fixed by `prompt.replace('\ud835', '')`, but it should be fixed automatically

I see in the readme they recommend to use mixed precision https://github.com/aqlaboratory/openfold/blob/3c1fd31ac47c8da54d088badd9eba61fe0b3fd26/docs/source/Training_OpenFold.md?plain=1#L129 It looks like a GPU setup problem, did you try the suggestion from the error? > If reserved...

I think it's still correct, because when it's called the order is reversed https://github.com/aqlaboratory/openfold/blob/3c1fd31ac47c8da54d088badd9eba61fe0b3fd26/openfold/data/data_transforms.py#L1109-L1114 Thought I agree it's confusing, maybe someone can confirm and we can update the docs?

What command do you run to train? That looks like a corrupted checkpoint error

Hi Trevor, as I understand, OpenFold uses [PyTorch dropout](https://github.com/aqlaboratory/openfold/blob/3c1fd31ac47c8da54d088badd9eba61fe0b3fd26/openfold/model/dropout.py#L44), and at the [PyTorch docs](https://pytorch.org/docs/stable/generated/torch.nn.Dropout.html) you can see that during eval mode it just computes an identity function. So you don't...

Please take a look @ricki-z @thepapermen