CUDA version is installed in cpu only envs #37
Comments
|
Ok this works: $ mamba install "openmm-torch=1.1=cpu*" pytorch-cpu=2I guess the issue is then that the cpu build is not being chosen automatically. |
|
Can you do |
$ mamba info -a
mamba version : 1.4.9
active environment : temp
active env location : /home/raul/mambaforge/envs/temp
shell level : 2
user config file : /home/raul/.condarc
populated config files : /home/raul/mambaforge/.condarc
/home/raul/.condarc
conda version : 22.9.0
conda-build version : not installed
python version : 3.10.8.final.0
virtual packages : __linux=6.4.12=0
__glibc=2.37=0
__unix=0=0
__archspec=1=x86_64
base environment : /home/raul/mambaforge (writable)
conda av data dir : /home/raul/mambaforge/etc/conda
conda av metadata url : None
channel URLs : https://conda.anaconda.org/conda-forge/linux-64
https://conda.anaconda.org/conda-forge/noarch
package cache : /home/raul/mambaforge/pkgs
/home/raul/.conda/pkgs
envs directories : /home/raul/mambaforge/envs
/home/raul/.conda/envs
platform : linux-64
user-agent : conda/22.9.0 requests/2.28.1 CPython/3.10.8 Linux/6.4.12-200.fc38.x86_64 fedora/38 glibc/2.37
UID:GID : 1000:1000
netrc file : None
offline mode : False
# conda environments:
#
base /home/raul/mambaforge
temp * /home/raul/mambaforge/envs/temp
sys.version: 3.10.8 | packaged by conda-forge | (main...
sys.prefix: /home/raul/mambaforge
sys.executable: /home/raul/mambaforge/bin/python
conda location: /home/raul/mambaforge/lib/python3.10/site-packages/conda
conda-build: None
conda-env: /home/raul/mambaforge/bin/conda-env
user site dirs: ~/.local/lib/python3.11
CIO_TEST: <not set>
CONDA_DEFAULT_ENV: temp
CONDA_EXE: /home/raul/mambaforge/bin/conda
CONDA_PREFIX: /home/raul/mambaforge/envs/temp
CONDA_PREFIX_1: /home/raul/mambaforge
CONDA_PROMPT_MODIFIER: (temp)
CONDA_PYTHON_EXE: /home/raul/mambaforge/bin/python
CONDA_ROOT: /home/raul/mambaforge
CONDA_SHLVL: 2
CPATH: <not set>
CURL_CA_BUNDLE: <not set>
LD_LIBRARY_PATH: /usr/local/cuda/lib64:
LIBRARY_PATH: /usr/local/cuda/lib64:
MANPATH: /usr/share/lmod/lmod/share/man:
MODULEPATH: /etc/modulefiles:/usr/share/modulefiles:/usr/share/modulefiles/Linux:/usr/share/modulefiles/Core:/usr/share/lmod/lmod/modulefiles/Core
MOZ_GMP_PATH: /usr/lib64/mozilla/plugins/gmp-gmpopenh264/system-installed
PATH: /home/raul/mambaforge/envs/temp/bin:/home/raul/mambaforge/condabin:/usr/local/bin:/usr/local/sbin:/usr/bin:/usr/sbin
REQUESTS_CA_BUNDLE: <not set>
SSL_CERT_FILE: <not set> |
|
Hmmm so this Which looks right, both Looking at mamba now, I am not sure why it is pulling in the GPU build. |
|
tl;dr |
|
Might be a mamba bug then? |
|
In which case? And yes I do think this is a mamba bug... I mean we may be doing something slightly wrong since if you just use |
|
I think this would be fixed if the CUDA build had a The reason mamba installs the CUDA version over the cpu version (on my machine) is simply because the timestamp of the CUDA build is newer: |
|
I believe this is the same issue detected here openmm/openmm-ml#67 |
|
The original issue has been solved by #49. Running the following succeeds in installing the cpu versions of everything: $ mamba create -n test openmm-torch pytorch-cpu
Looking for: ['openmm-torch', 'pytorch-cpu']
warning libmamba Cache file "/shared/raul/mambaforge/pkgs/cache/497deca9.json" was modified by another program
warning libmamba Cache file "/shared/raul/mambaforge/pkgs/cache/09cdf8bf.json" was modified by another program
conda-forge/noarch 13.4MB @ 32.3MB/s 1.2s
conda-forge/linux-64 32.2MB @ 22.2MB/s 7.0s
Transaction
Prefix: /shared/raul/mambaforge/envs/test
Updating specs:
- openmm-torch
- pytorch-cpu
Package Version Build Channel Size
────────────────────────────────────────────────────────────────────────────────────────
Install:
────────────────────────────────────────────────────────────────────────────────────────
...
+ pytorch 2.1.2 cpu_generic_py310h5d8fa8e_1 conda-forge 26MB
+ pytorch-cpu 2.1.2 cpu_generic_py310h9d11763_1 conda-forge 22kB
+ openmm-torch 1.4 cpu_py310h9717ab3_3 conda-forge 216kB |
Solution to issue cannot be found in the documentation.
Issue
I cannot manage to install the cpu only build of openmm-torch 1.1. On a machine without a GPU nor CUDA installation:
This results in an unusable library
Installed packages
Environment info
The text was updated successfully, but these errors were encountered: