Haotian Cui

Hi @litxiaoyao , we didn't filetered by is_primary_data when we trained the model. You can find the code for downloading data in the data forlder. The reason was just we...

Hi, thank you for the question. @ChloeXWang , could you please help with this?

Hi, thank you for the interest in our work! You can try newer version of python. In the meantime, flash-attn is an optional dependency now. Especially if you work with...

Hi, you can use the default settings in the tutorials. We use continuous MLP as binned value encoder. In the paper, we also stated as "We use fully connected layers,...

Oh, do you mean it should be like equation 13 in the sense that there should be a subscript j? Yeah, I think that can be more accurate. Are you...

Hi, we haven't tried that. I think it can make more sense to directly do a fresh training based on the code and model architecture. Loading the pre-trained human model...

Hi, @ChloeXWang could you please help with this?

Thank you @LenisLin and @moritzschaefer for your help. Yes, it is possible to run without flash-attn especially if you are using pytorch>=2.1. Can you let me know your specific needs,...

Hi, if you want to use the model in the zero-shot way, you may see this tutorial https://github.com/bowang-lab/scGPT/blob/main/tutorials/Tutorial_Reference_Mapping.ipynb. In summary, the notebook search for nearest neighbors based on cell embeddings,...

Hi, thank you for the question and interest in our work! You can find scripts to generate the figuers we reported here https://zenodo.org/records/10466117, also in the tutorial folder. I think...