Default pip install gives incompatible package versions

[james-vincent]

ImportError: cannot import name 'SparseDataset' from 'anndata._core.sparse_dataset'
[.....] python3.12/site-packages/anndata/_core/sparse_dataset.py

$ pip freeze |grep anndata
anndata==0.12.2
$ pip freeze |grep scvi
scvi-tools==0.20.3

[josephrich98]

I used the following steps to install and got a working environment (CUDA 12.2)

conda create -n scgpt python=3.9 -y
conda activate scgpt
conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=12.1 -c pytorch -c nvidia -y
pip install "numpy<2" "pandas>=1.3.5" "scvi-tools>=0.16.0,<1.0" "scanpy>=1.9.1,<2.0" "torchtext==0.16.2" "numba>=0.55.1" "scikit-misc>=0.1.4" "umap-learn>=0.5.3" "leidenalg>=0.8.10" "datasets>=2.3.0,<3.0" "typing-extensions>=4.2.0,<5.0" "scib>=1.0.3,<2.0" "cell-gears<0.0.3" "orbax<0.1.8" wandb
pip install scgpt --no-deps
pip install "flash-attn<1.0.5" --no-build-isolation
(optional) pip install "ipython==8.12.3" "jupyter-client<8" "prompt-toolkit<3.0.44"

[rivzzzzz]

Got installation to work on jupyterhub using the following (cuda 12.1, python 3.11.9, torch 2.2.1, numpy 1.26.4):

import sys
!{sys.executable} -m pip install --user --upgrade pip setuptools wheel
!{sys.executable} -m pip install --user scvi-tools==0.20.3 scanpy==1.9.3 anndata==0.9.2 scib==1.1.7
!{sys.executable} -m pip install --user torchtext==0.17.0 torchdata==0.7.1 --no-deps
!{sys.executable} -m pip install --user "numba>=0.55.1" "scikit-misc>=0.1.4" "umap-learn>=0.5.3" "leidenalg>=0.8.10" "datasets>=2.3.0,<3.0" "typing-extensions>=4.2.0,<5.0" "cell-gears<0.0.3" "orbax<0.1.8"
!{sys.executable} -m pip install --user --no-deps torch-scatter torch-sparse torch-cluster torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-2.2.1+cu121.html --no-cache-dir
!{sys.executable} -m pip install --user scgpt --no-deps
!{sys.executable} -m pip install --user wandb faiss-cpu nbformat

[angblabs]

I used the following steps to install and got a working environment (CUDA 12.2)

conda create -n scgpt python=3.9 -y
conda activate scgpt
conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=12.1 -c pytorch -c nvidia -y
pip install "numpy<2" "pandas>=1.3.5" "scvi-tools>=0.16.0,<1.0" "scanpy>=1.9.1,<2.0" "torchtext==0.16.2" "numba>=0.55.1" "scikit-misc>=0.1.4" "umap-learn>=0.5.3" "leidenalg>=0.8.10" "datasets>=2.3.0,<3.0" "typing-extensions>=4.2.0,<5.0" "scib>=1.0.3,<2.0" "cell-gears<0.0.3" "orbax<0.1.8" wandb
pip install scgpt --no-deps
pip install "flash-attn<1.0.5" --no-build-isolation
(optional) pip install "ipython==8.12.3" "jupyter-client<8" "prompt-toolkit<3.0.44"

Thank you so much @josephrich98! finally got scgpt working.

[lyr-sc]

我按照以下步骤进行安装，并获得了可运行的环境（CUDA 12.2）。

conda create -n scgpt python=3.9 -y
conda activate scgpt
conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=12.1 -c pytorch -c nvidia -y
pip install "numpy<2" "pandas>=1.3.5" "scvi-tools>=0.16.0,<1.0" "scanpy>=1.9.1,<2.0" "torchtext==0.16.2" "numba>=0.55.1" "scikit-misc>=0.1.4" "umap-learn>=0.5.3" "leidenalg>=0.8.10" "datasets>=2.3.0,<3.0" "typing-extensions>=4.2.0,<5.0" "scib>=1.0.3,<2.0" "cell-gears<0.0.3" "orbax<0.1.8" wandb
pip install scgpt --no-deps
pip install "flash-attn<1.0.5" --no-build-isolation
(optional) pip install "ipython==8.12.3" "jupyter-client<8" "prompt-toolkit<3.0.44"

Hello, after installing using your installation commands, I encountered an error when importing packages for fine-tuning the pre-trained model for cell type annotation in the tutorial:

'ImportError: cannot import name 'AlignedViewMixin' from 'anndata._core.aligned_mapping' (/root/miniconda3/envs/scgpt/lib/python3.9/site-packages/anndata/_core/aligned_mapping.py)'

I would like to ask how you solved this problem?