Dimensionality

#5
by midesk - opened

How do I instruct it to use a certain dimensionality? The default is 768.

As a workaround, the following code may be used to reduce the dimensionality from the default response:

def normalize_l2(x):
    x = np.array(x)
    if x.ndim == 1:
        norm = np.linalg.norm(x)
        if norm == 0:
            return x
        return x / norm
    else:
        norm = np.linalg.norm(x, 2, axis=1, keepdims=True)
        return np.where(norm == 0, x, x / norm)

Source: https://platform.openai.com/docs/guides/embeddings/use-cases
Chapter: Reducing embedding dimensions

It would be great if the GGUF version was capable of returning lower dimensions to improve the speed for certain use-cases.

Sign up or log in to comment