-
Notifications
You must be signed in to change notification settings - Fork 9
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Error:CUDA error: no kernal image is available for executio #87
Comments
Same as above Setup: Flame 2024 on rocky8.7, A5000 GPU (driver 525.116.04, cuda 12.0) |
Hi guys, this most likely comes from the fact that I'm trying to keep 2023 still supported and pytorch version 1.12.1 is too old to play well with your much more recent driver / gpu setup. I might have to update python 3.10 (Flame 2024) bundled pytorch to more recent version. In the meantime could you please try to whack it in manually and see if it works?
Please replace flame version with yours. That would create a folder named torchenv with python virtual environment from your flame python install
This will install numpy package into torchenv. You need to have internec connection open for this. 4)./torchenv/bin/python -m pip install torch==1.13.1+cu117 --extra-index-url https://download.pytorch.org/whl/cu117 This will install torch 1.13.1 for cuda 11.7. You can also try the most recent pytorch just by ./torchenv/bin/python -m pip install torch but I think trying out 1.13.1 is more safe approach remove everything from <your_flame_python_dir>/flameTimewarpML/packages/.lib/python3.10/site-packages/ This should upgrade pytorch that bundled with flameTimewarpML to 1.13.1 Let me know if this works |
Thank you. I will do as you ask. I appreciate it.
…On Fri, Dec 15, 2023 at 2:58 AM Andriy Toloshny ***@***.***> wrote:
Hi guys, this most likely comes from the fact that I'm trying to keep 2023
still supported and pytorch version 1.12.1 is too old to play well with
your much more recent driver / gpu setup. I might have to update python
3.10 (Flame 2024) bundled pytorch to more recent version.
In the meantime could you please try to whack it in manually and see if it
works?
In order to not to interfere with anything else the best would be to
create venv, get pytorch installed there and copy it over to
flameTimewarpML.
1. Navgate somewhere you have full write acces (home folder for
example)
2. /opt/Autodesk/python/2023.3/bin/python -m venv torchenv
Please replace flame version with yours. That would create a folder named
torchenv with python virtual environment from your flame python install
1. ./torchenv/bin/python -m pip install numpy
This will install numpy package into torchenv. You need to have internec
connection open for this.
4)./torchenv/bin/python -m pip install torch==1.13.1+cu117
--extra-index-url https://download.pytorch.org/whl/cu117
This will install torch 1.13.1 for cuda 11.7. You can also try the most
recent pytorch just by ./torchenv/bin/python -m pip install torch but I
think trying out 1.13.1 is more safe approach
remove everything from
<your_flame_python_dir>/flameTimewarpML/packages/.lib/python3.10/site-packages/
5) copy the contents of torchenv/lib/python3.9/site-packages/ to
<your_flame_python_dir>/flameTimewarpML/flameTimewarpML/packages/.lib/python3.10/site-packages/
This should upgrade pytorch that bundled with flameTimewarpML to 1.13.1
Let me know if this works
—
Reply to this email directly, view it on GitHub
<#87 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/BEV7HWNMJ424SSW54DVMEDLYJQUNDAVCNFSM6AAAAABAVSCJUOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNJXGY4DKOBZGQ>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
Thanks Talosh. Works perfectly with your instructions. Only thing of note is python3.9 is still the default on vanilla flame2024 rocky8.5 build. |
Anyone experience this or seen. this? Rocky8.7 Package Version numpy 1.24.1 [notice] A new release of pip is available: 23.0.1 -> 23.3.2
NVIDIA-SMI 525.116.04 Driver Version: 525.116.04 CUDA Version: 12.0 |
Thank you for pointing it out, I think it has been changed to 3.10 at some point in 2024 versions but need to double-check it |
First attempt running v0.5.0 dev 005. No picture preview and renders without picture too. (error below)
flameTimewarpML:Error:CUDA error: no kernal image is available for execution on the deviceCUDA kernal errors might be asynchronously reported as some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Setup: Flame 2024 on rocky8.5, A6000 GPU (driver 525.89.02, cuda 12.0)
Have updated PyTorch with no joy.
Shell log attached
Uploading flame2024_flame07_shell.log…
The text was updated successfully, but these errors were encountered: