Control: severity -1 normal The workaround was to reduce the tests run to those requiring less than 4GB memory. Anything larger was being too punishing on CPU-only tests anyway.
On 2026-01-13 23:21, Christian Kastner wrote: > Package: llama.cpp > Version: 5882+dfsg-4 > Severity: serious > > The llama.cpp autopkgtests currently fail if they are run on a GPU with > insufficient memory for the used models. The test runner must discover > this, and skip the tests for those models.
