diff --git a/services/hermes/switchyard-configmap.yaml b/services/hermes/switchyard-configmap.yaml index b953dfcb1..ad438875c 100644 --- a/services/hermes/switchyard-configmap.yaml +++ b/services/hermes/switchyard-configmap.yaml @@ -79,7 +79,7 @@ data: max_retries = 0 [targets.classifier] - id = "qwen2.5:3b-instruct-q4_0" + id = "qwen2.5:14b-instruct-q4_0" llm_client = "classifier" [targets.local_qwen_low] diff --git a/services/hermes/switchyard-deployment.yaml b/services/hermes/switchyard-deployment.yaml index 71bf2c117..0d23c8097 100644 --- a/services/hermes/switchyard-deployment.yaml +++ b/services/hermes/switchyard-deployment.yaml @@ -19,7 +19,7 @@ spec: labels: app: hermes-switchyard annotations: - ai.bstein.dev/config-rev: "20260811-switchyard-authority-v7" + ai.bstein.dev/config-rev: "20260811-switchyard-authority-v8" prometheus.io/scrape: "true" prometheus.io/port: "9005" prometheus.io/path: /metrics diff --git a/testing/tests/test_hermes_chat_quality.py b/testing/tests/test_hermes_chat_quality.py index 9fcee3d6d..780cc68b1 100644 --- a/testing/tests/test_hermes_chat_quality.py +++ b/testing/tests/test_hermes_chat_quality.py @@ -766,6 +766,8 @@ def test_local_flux_runtime_and_gpu_handoff_are_flux_managed(): switchyard = _documents(HERMES / "switchyard-configmap.yaml")[0]["data"][ "routes.toml" ] + assert 'id = "qwen2.5:14b-instruct-q4_0"' in switchyard + assert "qwen2.5:3b-instruct-q4_0" not in switchyard assert "route/local/qwen2.5-14b/medium" in switchyard model_gate = _documents(HERMES / "model-gate-configmap.yaml")[0]["data"][ "model_gate.py" @@ -774,7 +776,7 @@ def test_local_flux_runtime_and_gpu_handoff_are_flux_managed(): def test_titan20_serializes_classifier_and_local_chat_model_residency(): - """The two Qwen weights must not overcommit Xavier unified memory.""" + """Classifier and local chat share one serialized resident Qwen weight.""" deployment = _documents( Path(__file__).parents[2] / "services/ai-llm/deployment.yaml" )[0]