In-Context Learning on Llama3-8B Scenario 3: More layers (L={0,4,8,...})