This app uses ModelJars 0.1.6, Models 0.2.6, and the qualified Qwen3 0.6B Q4_0
marker. The runtime selects the qualification-owned chat template, reports and verifies
the cache on first use, exercises the exposed inference pipeline, runs inference
in-process, checks the answer, and prints
PLAIN_JAVA_E2E_PASS.
This is a lightweight onboarding and integration demo for one known model. It is not a catalog-wide model/backend compatibility test and is not a release qualification gate.
./run.shAn opt-in Gemma 4 compatibility probe uses the 15.64 GiB Q4_K_M artifact and an 8 GiB Java heap:
./run.sh gemma4That profile exists to exercise the large-model path explicitly; Qwen remains the fast default. Gemma 4 must pass the repository's default-configuration qualification smoke before it is presented as qualified again.
Requirements: Maven and JDK 25. The script uses the local SDKMAN Temurin 25 installation
when the current JAVA_HOME is older.