Missing libjllama on Linux aarch64
AutonomicPerfectionist opened this issue · 3 comments
I have not had the time to test the arm variant yet but I took a look at the generated maven artifact and noticed the aarch64 platform is missing the libjllama.so library, I assume this would cause problems
Appears the latest workflow run silently failed during compilation, a syntax error in llama.cpp's common/common.cpp
Might be an upstream problem, but at the very least the workflow should probably be adjusted to correctly fail on compilation issues
Just adding that it's the same issue for Mac as well. aarch64 is missing on the latest release.
You're right, both macos and linux aarch64 files are incomplete. I added some unit tests that should be run during the workflow to validate everything works. This doesn't work yet, since I need a model file in the github runner.
Maybe a model could be downloaded for example from huggingface hub. But I think I saw some repos that had tiny testing models. Those could be directly added to the repository. Another solution might be to add a minimal test that just tries to load the library. I'll find a solution for the next release.