SmoothQuant Scaled Weights

Problem

Apply SmoothQuant channel scaling to both weights and activations. Multiply each weight column by its smoothing factor s_j and divide each activation column by the same factor: W_scaled[i][j] = W[i][j]*s_j and X_scaled[i][j] = X[i][j]/s_j. Return [W_scaled, X_scaled].

The signature is smoothquant_scaled_weights(W, X, s).

Hint

The product W_scaled * X_scaled equals W * X, so the layer output is unchanged.

Test cases

case 1

in: [[[1,2]],[[2,4]],[2,4]]

expected: [[[2,8]],[[1,1]]]

case 2

in: [[[1,1],[1,1]],[[1,1]],[0.5,2]]

expected: [[[0.5,2],[0.5,2]],[[2,0.5]]]

case 3

in: [[[2,4]],[[1,1]],[1,1]]

expected: [[[2,4]],[[1,1]]]

Starter codePython

def smoothquant_scaled_weights(W, X, s):
    # Your code here
    pass

3 test cases · run your code against them in the editor