Test Code Generation with Performance Constraints

Describe a test case where your agent is tasked with generating a custom PyTorch layer (e.g., a specific type of activation function) given high-level requirements and explicit performance constraints (e.g., 'must be low memory footprint'). Detail how the agent will use its RAG capabilities to find relevant examples and then leverage DeepSeek-R1 to generate the code, followed by validation using your Marvin-MCP integrated tools (e.g., a mock memory usage checker).