feat(agent): persist production generation settings

ai_knowledge_base_v2/apps/backend/tests/test_agent_runtime_config.py

@@ -0,0 +1,101 @@
+from decimal import Decimal
+
+from sqlalchemy import create_engine
+from sqlalchemy.orm import Session
+from sqlalchemy.pool import StaticPool
+
+from app.api.admin_settings import get_agent_runtime_config, save_agent_runtime_config
+from app.models import Base
+from app.models.admin import Admin
+from app.models.ai_config import ModelConfig
+from app.schemas.admin import AgentRuntimeConfigSaveRequest
+from app.services.model_stream_service import _openai_stream_payload
+from app.services.model_service import _max_output_tokens
+from app.services.rag_service import RagResult
+
+
+def _database() -> Session:
+    engine = create_engine(
+        "sqlite:///:memory:",
+        connect_args={"check_same_thread": False},
+        poolclass=StaticPool,
+    )
+    Base.metadata.create_all(engine)
+    return Session(engine)
+
+
+def _admin() -> Admin:
+    return Admin(id=1, username="admin", password="hash", name="系统管理员", status=1)
+
+
+def _model() -> ModelConfig:
+    return ModelConfig(
+        id=1,
+        provider="openai-compatible",
+        display_name="正式模型",
+        api_type="openai_compatible",
+        model_name="production-model",
+        base_url="https://example.com/v1",
+        api_url="",
+        api_key="encrypted",
+        auth_type="bearer",
+        max_token=None,
+        stream_enabled=1,
+        timeout_second=30,
+        enabled=1,
+    )
+
+
+def test_runtime_config_defaults_to_long_answer_safe_max_tokens():
+    with _database() as db:
+        admin = _admin()
+        model = _model()
+        db.add_all([admin, model])
+        db.commit()
+
+        result = get_agent_runtime_config(db=db, current_admin=admin)["data"]
+
+        assert result["modelId"] == model.id
+        assert result["modelName"] == "正式模型"
+        assert result["maxToken"] == 8192
+        assert _max_output_tokens(model) == 8192
+
+
+def test_saved_runtime_config_is_persisted_on_enabled_model():
+    with _database() as db:
+        admin = _admin()
+        model = _model()
+        db.add_all([admin, model])
+        db.commit()
+
+        result = save_agent_runtime_config(
+            AgentRuntimeConfigSaveRequest(
+                temperature=0.3,
+                topP=0.9,
+                topK=40,
+                presencePenalty=0.2,
+                frequencyPenalty=0.4,
+                maxToken=12000,
+            ),
+            db=db,
+            current_admin=admin,
+        )["data"]
+
+        db.refresh(model)
+        assert result["maxToken"] == 12000
+        assert model.max_token == 12000
+        assert model.temperature == Decimal("0.30")
+        assert model.top_p == Decimal("0.90")
+        assert model.top_k == 40
+        assert model.presence_penalty == Decimal("0.20")
+        assert model.frequency_penalty == Decimal("0.40")
+
+        payload = _openai_stream_payload(
+            model,
+            RagResult(question="测试", knowledge_scopes=[], chunks=[], prompt="测试", allow_general_knowledge=True),
+        )
+        assert payload["max_tokens"] == 12000
+        assert payload["temperature"] == 0.3
+        assert payload["top_p"] == 0.9
+        assert payload["presence_penalty"] == 0.2
+        assert payload["frequency_penalty"] == 0.4