I’m trying to install and run vllm & several of the dependencies require pytorch 2.8.0, however no matter what I try the most recent version I can install is 2.6.0 (updating only goes to 2.6 and after trying to specify version, pip returns there is no version 2.8). For various reasons building from source (per the git page) is not an option. Any suggestions appreciated!
Could you post your OS, Python version, CPU arch, please?
Yes sorry -
OS - CentOS Linux release 7.4.1708 (Core)
Python 3.11.4
CPU x86_64
Thank you!
I cannot reproduce any issues and can install PyTorch 2.8.0 with CUDA 12.8 using pip install torch as given in our install instructions.
lscpu | grep -i arch
Architecture: x86_64
python --version
Python 3.11.13
pip install torch
Collecting torch
Downloading torch-2.8.0-cp311-cp311-manylinux_2_28_x86_64.whl.metadata (30 kB)
...
I don’t know what might be causing the issue in your setup, but you could try to manually download the wheel and pip install it from your local folder. Maybe the error message would be better in this case.
Yeah - I have an old version of torch I can’t remove (working on an HPC so some files are blocked) but when I try to upgrade torch it defaults to 2.6.0:
pip install -U torch
Defaulting to user installation because normal site-packages is not writeable
Requirement already satisfied: torch in /share/apps/centos7/anaconda3/2023.07/lib/python3.11/site-packages (2.0.1)
Collecting torch
Downloading torch-2.6.0-cp311-cp311-manylinux1_x86_64.whl.metadata (28 kB)
```