Your question is Implement Attention and Complexity. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
Explain how you would implement multi-head self-attention from scratch in PyTorch, including the time and memory complexity, and then describe how FlashAttention reduces memory usage.