CodeBench: mallikohtaiset promptiprofiilit (profiles.json)

- profiles.json: malli → profiili → prompti -mappaus
- code-small.md: tiivistetty prompti pienille malleille (8b, 4b)
- benchmark valitsee automaattisesti oikean promptin mallin perusteella
- qwen3-coder:30b → code.md (large), qwen3:8b → code-small.md (small)

kipina-codebench/prompts/code-small.md

@@ -0,0 +1,22 @@
+Generate a FastAPI project with SQLAlchemy and SQLite. Follow the REFERENCE IMPLEMENTATION exactly.
+
+Generate these 4 files with === markers:
+
+=== models.py ===
+=== schemas.py ===
+=== main.py ===
+=== test_main.py ===
+
+Key patterns (copy from reference):
+- class Base(DeclarativeBase): pass
+- Mapped[str] = mapped_column(String(255))
+- Mapped[str | None] = mapped_column(Text, default=None)
+- model_config = ConfigDict(from_attributes=True)
+- model_dump() not dict()
+- POST 201, GET list, GET by id 404, PUT, DELETE 204
+
+CRITICAL:
+- Use ONLY fields from the JSON spec — no created_at or extra fields
+- Generate EXACTLY 6 tests per entity: create, list, get_by_id, not_found, update, delete
+- No search, filter, or other extra tests
+- No markdown fences in output