I get the "good" result with phi-4 and gemma-3n in RAG scenario - i.e. it only used context provided to answer and couldn't answer questions if context lacked t

I get the "good" result with phi-4 and gemma-3n in RAG scenario - i.e. it only used context provided to answer and couldn't answer questions if context lacked the answer without hallucination.