$\mu$ LO: Compute-Efficient M | Pangram Labs