Unable to install pytorch 2.10 using conda 26.3.2 on Windows 11

windows
1

Hi

Has anyone had issues installing pytorch using conda? We have an on-premises Anacoda repo (air gapped /w no Internet) and I am unable to install pytorch using ‘conda install pytorch’ in a clean environment.

conda create -n pytorch

conda activate pytorch

conda clean --all

conda install pytorch (or pytorch-cpu)

I get this error

>>>

conda.CondaMultiError: Error with archive C:\conda_pkgs\pytorch-2.10.0-cpu_mkl_py314hf6a6a2c_103.conda. You probably n:
failed with error: [Errno 2] No such file or directory: ‘C:\conda_pkgs\pytorch-2.10.0-cpu_mkl_py314hf6a6a2c_103\info’

>>>

>>>

Downloading and Extracting Packages:

InvalidArchiveError("Error with archive C:\conda_pkgs\pytorch-2.10.0-cpu_mkl_py314hf6a6a2c_103.conda. You probably need to delete and re-download or re-create this file. Message was:\n\nfailed with error: [Errno 2] )

>>>

I can install all package dependencies except for pytorch itself. Other users have installed it so I am not sure if the .conda file itself might be corrupt in the repo. Installing other versions of pytorch < 2.10 results in the same error.

Reinstalling conda did not fix the problem.

The version in the Anaconda repo is pytorch(-cpu) v2.10.0 and was last updated May 6, 2026

Any suggestions?

Take care,

Shayne

2

I don’t know which conda channel you are using but please note that we are not publishing the PyTorch binary on conda anymore as announced here.

3

Thanks, that would be the cause of the problem. Unfortunately I am using a secure air-gapped Analytics VM with no Internet connection and can only install curated packages from the local conda repo. Might there be any pytorch alternatives that are compatible with conda?

Take care,

Shayne

4

You could try to install the conda-forge builds but also keep in mind that these are community-maintained builds and I don’t know how well tested these are. E.g. I cannot find the latest PyTorch 2.11 release there so am unsure if this release is still being planned for conda-forge.

You can install any pip wheels into conda environments in case this works as a lot of users are doing this for years without any issues (and without installing any conda libs at all).