feat(benchmark): add lme_one_question job to jinli_lme.yaml

reme/config/jinli_lme.yaml

@@ -92,6 +92,41 @@ jobs:
     steps:
       - backend: python_execute_step
 
+  lme_one_question:
+    backend: base
+    description: "Run context_answer_step + answer_judge_step for a single LME question."
+    parameters:
+      type: object
+      properties:
+        query:
+          type: string
+          description: "the LME question being answered"
+        session_context:
+          type: string
+          description: "concatenated chat history for the relevant session"
+        current_date:
+          type: string
+          description: "today's date in YYYY-MM-DD, used to resolve relative phrases"
+        agent_answer:
+          type: string
+          description: "answer from the agent-under-test; can be the value of `context_answer` from a prior call"
+        golden_answer:
+          type: string
+          description: "ground-truth answer from the LME dataset"
+        question_type:
+          type: string
+          description: "LME category; one of temporal_reasoning / knowledge_update / single_session_preference / other"
+      required:
+        - query
+        - session_context
+        - current_date
+        - agent_answer
+        - golden_answer
+        - question_type
+    steps:
+      - backend: context_answer_step
+      - backend: answer_judge_step
+
 components:
   tokenizer:
     default:

tests/unit/test_config_parser.py

@@ -86,3 +86,27 @@ def test_expand_env_vars_converts_expanded_scalar_types(monkeypatch):
         "url": "http://localhost:18080",
         "string_bool": "false",
     }
+
+
+def test_jinli_lme_config_registers_lme_one_question_job():
+    """``jinli_lme.yaml`` exposes a ``lme_one_question`` job that bundles the
+    LME ``context_answer`` and ``answer_judge`` steps and requires every
+    input the two steps consume."""
+    cfg = _load_config("jinli_lme.yaml")
+
+    job = cfg["jobs"]["lme_one_question"]
+    assert job["backend"] == "base"
+    assert [step["backend"] for step in job["steps"]] == [
+        "context_answer_step",
+        "answer_judge_step",
+    ]
+    assert job["parameters"]["required"] == [
+        "query",
+        "session_context",
+        "current_date",
+        "agent_answer",
+        "golden_answer",
+        "question_type",
+    ]
+    props = set(job["parameters"]["properties"])
+    assert props == set(job["parameters"]["required"])