This is inspired by #66. For very small models where each layer produces a short feature vector (say ≤ O(100)) then we could explore using Static arrays for everything i.e. keep the entire chain on the stack. This might lead to extremely fast models.
This is inspired by #66. For very small models where each layer produces a short feature vector (say ≤ O(100)) then we could explore using Static arrays for everything i.e. keep the entire chain on the stack. This might lead to extremely fast models.