Accessing Attention on MLX

[tburleyinfo]

“How can intermediate attention activations (e.g., attention heads or attention weights) be accessed or inspected when running transformer models with MLX?”