-
Notifications
You must be signed in to change notification settings - Fork 937
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
struggling compiling alphazero_torch version #872
Comments
On first glance it seems like you are not linking in some abseil targets. Possibly the linker is picking up the binary abseil libraries from a different location which are missing those symbols (or you're just not linking them at all for some reason?) Abseil would be downloaded and installed when you run We run the AlphaZero torch test every time a new PR is submitted and every time we sync to github each week, and it's run on Ubuntu 22.04 in this CI target here: https://github.com/deepmind/open_spiel/blob/6a49b935bf674920071d0a23464a6d4fa3b39618/.github/workflows/actions.yml#L15 I expect it can be fixed by settings some paths differently. I'll respond more after you answer the first few questions above. |
I noticed tensorflow in the Docker image name. I believe Tensorflow includes past versions of abseil in it, so it's possible that your linker is finding and linking against those, rather than the ones included in OpenSpiel. Can you inspect the output to see if it lists the path to the abseil library it's trying to link against? It might have something like |
Any luck? |
Hi, I also met the same problem.
First, I can build without libtorch setting (OPEN_SPIEL_BUILD_WITH_LIBTORCH='OFF' ... etc).
And yes, I have 'abseil-cpp' folder in '/build'. Thanks for reading this! |
Hello,
(and Kudos for the amazing work!)
I'm struggling since a couple of days to have a working version of alphazero libtorch version (python version works fine, but I guess it'll takes ages to train a model)
During compilation phase (/build_and_run_tests.sh), I get compilation errors on alpha_zero_torch_example or torch_integration_test like those :
Some context info:
(basically I modified the provided Dockerfile.base to include nvidia/cuda/cudnn stuff)
So my question is : would you have a reference configuration that enable usage of alphazero libtorch version ? (os|libtorch version | cuda ...)?
Many thanks in advance
Julien
The text was updated successfully, but these errors were encountered: