diff --git a/benchmark/mobilemem/_audit.py b/benchmark/mobilemem/_audit.py index 44dd394e..5bb1ba2f 100644 --- a/benchmark/mobilemem/_audit.py +++ b/benchmark/mobilemem/_audit.py @@ -166,13 +166,11 @@ def judge_settings(app) -> dict: fields = ("backend", "model", "stream", "context_size", "max_retries", "retry_delay") return { **{key: safe_value(model[key]) for key in fields if key in model}, - "parameters": safe_value( - { - key: value - for key, value in (model.get("parameters") or {}).items() - if key in {"temperature", "max_tokens", "thinking_enable", "top_p"} - }, - ), + "parameters": { + key: safe_value(value) + for key, value in (model.get("parameters") or {}).items() + if key in {"temperature", "max_tokens", "thinking_enable", "top_p"} + }, "extra_body": {"enable_thinking": (model.get("extra_body") or {}).get("enable_thinking")}, "client_kwargs": { key: value for key, value in (model.get("client_kwargs") or {}).items() if key in {"timeout", "max_retries"} diff --git a/tests/unit/test_mobilemem_benchmark.py b/tests/unit/test_mobilemem_benchmark.py index 3309eab8..3c7da30f 100644 --- a/tests/unit/test_mobilemem_benchmark.py +++ b/tests/unit/test_mobilemem_benchmark.py @@ -159,6 +159,10 @@ async def test_candidate_and_judge_start_with_native_settings(tmp_path, monkeypa audit = mobilemem.judge_settings(judge_app) assert audit["model"] == "offline-judge-override" assert audit["parameters"] == {"max_tokens": 4096, "temperature": 0, "thinking_enable": False} + with monkeypatch.context() as environment: + for field in ("temperature", "max_tokens", "thinking_enable"): + environment.setenv("TEST_PRIVATE_KEY", field) + assert mobilemem.judge_settings(judge_app)["parameters"] == audit["parameters"] assert audit["protocol"] == "mobilemem-omni-membase-default-judge-v1" assert "credential" not in json.dumps(audit) and "provider.invalid" not in json.dumps(audit) models = candidate_app.context.components["as_llm"]