Is there a way to download the embeddings of Evo2?

Question

[Jimmy]

I am trying to access the embeddings released in the new evo2 paper https://arcinstitute.org/manuscripts/Evo2

The embeddings should be available through their github: https://github.com/ArcInstitute/evo2/tree/main/evo2

However, the listed git clone command listed on the page is not working for me (a permission error occurs) and I do not have access to the necessary GPU resources to run evo2 anyways. Alternatively, I could try to download the embeddings by themselves, but my issue is that I cannot identify where on the GitHub the embeddings file is located. Is there any way I can download the evo2 embeddings by themselves from the GitHub (or potentially through another website/resource)?

Answer 1

[Mensur Dlakic]

I think you were trying to clone a page that is two levels below the main:

https://github.com/ArcInstitute/evo2

You should be able to get a clone link from that page. There are also links to pre-trained models on HuggingFace.

[Jimmy]

I included that link accidentally. I still have issues with the cloning. Using 'git clone' alone (and not the given command on the GitHub) can clone the repository, but the "pip install ." stage does not work then. Is it necessary to clone the entire repository to access their embeddings?

To use it through HuggingFace (say, evo_7b), would I have to download the 13.8 Gb file here? https://huggingface.co/arcinstitute/evo2_7b/tree/main

Apologies for the simple questions.

[Mensur Dlakic]

Using 'git clone' alone (and not the given command on the GitHub)

Everything is clearly explained on their web site. You can't expect to install the package correctly if you are not following the instructions. Without a GPU, you may need to do everything on their website.

To use it through HuggingFace (say, evo_7b), would I have to download the 13.8 Gb file here?

You will have to download one of the models, and evo2_1b_base seems to be the smallest.