chore(packaging): remove legacy evaluator wiring

TESTING.md

@@ -434,8 +434,7 @@ make test-package PACKAGE=nmp_common
 uv run pytest -v packages/nmp_common/tests/
 
 # Test a specific service
-make test-service SERVICE=evaluator
-uv run pytest -v services/evaluator/tests/
+make test-service SERVICE=guardrails
 
 # Test a specific file
 uv run pytest -v path/to/test_file.py
@@ -746,4 +745,3 @@ If you're updating existing tests:
 - [ ] Chaos engineering tests
 - [ ] Load and stress tests
 - [ ] Contract testing between services
-

conftest.py

@@ -232,7 +232,7 @@ def pytest_collection_modifyitems(config, items):
                 item.add_marker(pytest.mark.e2e)
                 marker_names.add("e2e")
 
-        # Auto-mark integration tests (e.g., /services/evaluator/tests/integration/)
+        # Auto-mark integration tests (e.g., /services/core/jobs/tests/integration/)
         elif "/integration/" in fspath_str:
             if "integration" not in marker_names:
                 item.add_marker(pytest.mark.integration)

docs/evaluator/tutorials/run-llm-judge-evaluation.mdx

@@ -155,12 +155,12 @@ This tutorial uses `nvidia/nemotron-3-nano-30b-a3b` from NVIDIA Build.
 ```python
 from nemo_evaluator_sdk import RunConfig, LLMJudgeMetric
 from nemo_evaluator_sdk.values import (
+    FilesetRef,
     InferenceParams,
     JSONScoreParser,
     Model,
     RangeScore,
 )
-from nmp.evaluator.app.values import FilesetRef
 
 JUDGE_MODEL_URL = "https://integrate.api.nvidia.com/v1/chat/completions"
 JUDGE_MODEL_NAME = "nvidia/nemotron-3-nano-30b-a3b"

docs/set-up/config-reference.mdx

@@ -813,44 +813,6 @@ unsloth:
   default_training_execution_profile: gpu
 ```
 
-### `evaluator`
-
-Configuration for the Evaluator service.
-
-```yaml
-evaluator:
-  # Configuration for jobs created with Evaluator service.
-  jobs:
-    # Directory path in job container for evaluation configuration. | default: '/configs'
-    configs_dir: /configs
-    # Directory path of the shared volume mount for job steps to persist artifacts for a job. | default: '/jobs'
-    volume_path: /jobs
-    # Directory path in the job container for results to be output. | default: '/jobs/results'
-    results_dir: /jobs/results
-    # Directory path in the job container for dataset files to be downloaded to and loaded from. | default: '/jobs/datasets'
-    dataset_dir: /jobs/datasets
-  # Configuration for EvalFactory integration with NeMo Platform.
-  evalfactory:
-    # default: 'nvcr.io/nvidia/eval-factory/agentic_eval:26.01'
-    agentic_eval: nvcr.io/nvidia/eval-factory/agentic_eval:26.01
-    # default: 'nvcr.io/nvidia/eval-factory/bfcl:26.01'
-    bfcl: nvcr.io/nvidia/eval-factory/bfcl:26.01
-    # default: 'nvcr.io/nvidia/eval-factory/lm-evaluation-harness:26.01'
-    lm_eval_harness: nvcr.io/nvidia/eval-factory/lm-evaluation-harness:26.01
-    # default: 'nvcr.io/nvidia/eval-factory/bigcode-evaluation-harness:26.01'
-    bigcode_evaluation_harness: nvcr.io/nvidia/eval-factory/bigcode-evaluation-harness:26.01
-    # default: 'nvcr.io/nvidia/eval-factory/rag_retriever_eval:26.01'
-    rag_retriever: nvcr.io/nvidia/eval-factory/rag_retriever_eval:26.01
-    # default: 'nvcr.io/nvidia/eval-factory/safety-harness:26.01'
-    safety_harness: nvcr.io/nvidia/eval-factory/safety-harness:26.01
-    # default: 'nvcr.io/nvidia/eval-factory/simple-evals:26.01'
-    simple_evals: nvcr.io/nvidia/eval-factory/simple-evals:26.01
-    # Connect to a hosted Milvus server for retrieval evaluations
-    milvus_url:
-  # Upsert system metrics and benchmarks on app startup | default: False
-  recreate_existing_system_entities: false
-```
-
 ### `safe_synthesizer`
 
 Configuration for Safe Synthesizer plugin API and task compilation.

e2e/test_inference.py

@@ -236,7 +236,6 @@ def test_model_list_via_openai_route(sdk: NeMoPlatform, workspace: str):
         workspace=workspace,
         name=entity_name,
         mock_response_body={"id": "chatcmpl-test", "choices": []},
-        served_models={entity_name: f"served-{entity_name}"},
     )
 
     models = sdk.inference.gateway.openai.v1.models.list(workspace=workspace)