Uh oh!
There was an error while loading. Please reload this page.
fix(IndexerJob,ScanService): Do not attempt to remove prefix from paths - #137
Conversation
it doesn't seem to work Signed-off-by: Marcel Klehr <mklehr@gmx.net>
kyteinsky
left a comment
There was a problem hiding this comment.
the test fail seems to be coming from the llama_cpp_python library, should not matter here.
marcelklehr
commented
Jul 4, 2025
Is that a bug in the test setup then or in the backend? |
kyteinsky
commented
Jul 4, 2025
Both in a way. The error comes from the backend and in the test setup, we install the python deps from the requirements.txt (https://github.com/nextcloud/context_chat/actions/runs/16073429868/workflow?pr=137#L178) file which is non-versioned. The versioning load is taken by the docker container where a particular version of llama_cpp_python pre-built package is installed and for the other packages, it the latest version. But yeah here we can't test it right now with this bug, might be worth to pin the package in ccb for a while. |
marcelklehr
commented
Jul 7, 2025
mmh, can we take a leaf out of the book of llm2? Cause it seems to work there? |
kyteinsky
commented
Jul 7, 2025
removing a dependency from the poetry is not straightforward though, it modifies the lock file. We need to do this to ship one lock file for the manual install in CI and remove it for the docker build. Even after we switch to building llama-cpp-python, it should be in a different build context in the docker container so we ship the lighter runtime image of cuda only, landing us in the same position. |
it doesn't seem to work