-
Notifications
You must be signed in to change notification settings - Fork 4
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
Do you think you'll ever migrate your code to the latest stable versions of mmcv and mmdetection? #7
Comments
Hi, I re-implemented the code with a newer CUDA without the need for changing the mmcv and mmdetection versions. I tried this code on an NVIDIA A100-SXM4-80GB with CUDA 11.6, building the entire code environment from scratch as follows:
Then pip run a requirements.txt file with the following content:
After Set your CUDA_HOME variable as:
This was the minimal setup using which I was able to run with latest CUDA versions, without changing the mmcv and mmdetection versions. Hope this helps you. |
Thanks, I was able to get further this time, but encountered problems when I tried to run Did you also get a whole list of warnings when you built yours? These are my gcc and nvcc versions, not sure what other tools I need to sync up (or if the gcc version matters at all, but I'm sure the nvcc version does matter): gcc (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0 Cuda compilation tools, release 11.6, V11.6.55 |
Yes @nbkn865 I also got several warnings like you said, but it was able to build successfully. As far as I remember, I also got some GCC-related warnings but I think gcc > 7 should work. This is the output of my collect_env:
The nvcc version looks like:
In this environment, my code does run. If it still doesn't work, I'm sure you will find something on StackOverflow. I also did the last time I ran but haven't bookmarked the link to how I did it. But yes, gcc-related issue was solved once I took in GCC-7. |
That's interesting, you used CUDA 11.4 for nvcc but loaded CUDA 11.6 for Torch: Let me try that CUDA version as well |
Actually, can you share the full printout of Mine is:
|
What's interesting is that the CUDA runtime is But shouldn't I think if you can share your collect_env printout, I can try to figure out how to replicate it as close as possible, and hopefully I can get your repo to work |
The NVIDIA A100 I am using comes with CUDA 11.4. Moreover, when I install cudatoolkit-dev, by default it gives me cudatoolkit-dev 11.4. That is why my nvcc has CUDA 11.4. However, I could not find cudatoolkit=11.4 for any pytorch version on PyTorch website, hence went with 11.6 for cudatoolkit. Are you sure you have installed cudatoolkit-dev? I think if you do install, then your cuda runtime version will be set to 11.4 like mine. Here is my complete output of collect_env for your reference:
NOTE: a couple library versions like numpy might have been upgraded/downgraded in this since I have made some changes to my conda environment for another work. |
Hi, I used the torch1.12.1 cuda11.6 but got the erro: Traceback (most recent call last): |
Hi, Can you tell me more specifically about your environment settings and the output of |
My environment settings are as follows:nvcc: NVIDIA (R) Cuda compiler driver packages in environment at /home/dell/anaconda3/envs/zsd_scratch:Name Version Build Channel_libgcc_mutex 0.1 main defaults |
@sqdcbbj Looks like there are some dependency issues since your nvcc has CUDA 11.5 but your environment contains cudatoolkit 9.0. You can try upgrading this by installing the cudatoolkit 11.x runtime following the steps given here. |
When executing this code, I encountered some problems {conda install - c conda forge cudatoolkit dev} ==>Script Output<== Stderr:% Total% Received% Xferd Average Speed Time Time Current F: Anaconda envs zsd bin cudatoolkit dev post install. py: 281: DeprecationWarning: distutils version class is deprecated. Please use packaging.version instead. ==>Script Output<== |
I think you mistyped the command a little bit - you missed some hyphens. It should be:
|
I closed out my other issue because I couldn't get the repo to work because of the old code, so was wondering/hoping if you plan to bring your code up to date with the latest mmcv and mmdetection packages
The text was updated successfully, but these errors were encountered: