LoRA Update

Compute the LoRA low-rank update of a layer's output: $B(Ax)B\backslash ,(A\backslash ,x)$, where A has shape (r, d_in), B has shape (d_out, r) and x is an input vector (or a matrix with one input per column). Return only the update, not the base layer output.

Implement solve(x, A, B).

Returns. Return a NumPy array of shape (d_out,) for a vector input.

Examples

Example 1

Input:

solve([1.0, 2.0, 3.0], [[1.0, 0.0, 0.0], [0.0, 1.0, 0.0]], [[1.0, 1.0], [2.0, 0.0]])

Output:

[3.0, 2.0]

Example 2

Input:

solve([1.0, 1.0], [[0.0, 0.0]], [[5.0], [6.0]])

Output:

[0.0, 0.0]