I want to build DL4J from source for my cuda 12.8 but it gives this error:
[ERROR] Failed to execute goal org.bytedeco:javacpp:1.5.11:build (javacpp-cppbuild-validate) on project libnd4j: Execution javacpp-cppbuild-validate of goal org.bytedeco:javacpp:1.5.11:build failed: Plugin org.bytedeco:javacpp:1.5.11 or one of its dependencies could not be resolved: The following artifacts could not be resolved: org.bytedeco:javacpp:jar:linux-x86_64-cuda:1.5.11, org.bytedeco:openblas:jar:linux-x86_64-cuda:0.3.28-1.5.11: Could not find artifact org.bytedeco:javacpp:jar:linux-x86_64-cuda:1.5.11 in central ( ) → [Help 1]
[ERROR]
I’m referencing from here: deeplearning4j/.github/workflows/build-deploy-linux-cuda-12.6.yml at master · deeplearning4j/deeplearning4j · GitHub
@manvendra-singh-2709 can you clarify if you want just M2.1 with cuda 12.x support or the new features? If you want the newer features then I would suggest waiting on snapshots or I can point you at my dev branch. I’m running cuda 12.x no issue. I can show you how to patch M2.1 with cuda 12.x if you just want that though. I’ll have a snapshot build in the next day or 2 if you want as well.
My main goal is CUDA 12.8 compatibility. I’m fine with either M2.1 patched or snapshot builds whichever you recommend as more stable. But I would definitely like to fry out the sanpshot builds as well.
@manvendra-singh-2709 I’m in the middle of a big merge to master as you can see from the various pull requests. This should work for you: Dev branch to be segment by agibsonccc · Pull Request #10229 · deeplearning4j/deeplearning4j · GitHub cuda 12.9 support is in there. You should just need to run mvn -Pcuda -Dlibnd4j.chip=cuda -Dlibnd4j.buildthreads=4 clean install -DskipTests
As needed do ./change-cuda-versions.sh
Note you WILL* need your environment setup correctly. Please do let me know if you have issues.
@manvendra-singh-2709 not sure what the issue with the forums was sorry about that. I’ve published linux snapshots now with cuda 12.9.