We present a comprehensive observational study of the gas phase metallicity of star-forming galaxies from z ~ 0 -> 3. We combine our new sample of gravitationally lensed galaxies with existing lensed and non-lensed samples to conduct a large investigation into the mass-metallicity (MZ) relation at z > 1. We apply a self-consistent metallicity calibration scheme to investigate the metallicity evolution of star-forming galaxies as a function of redshift. The lensing magnification ensures that our sample spans an unprecedented range of stellar mass (3*10^{7}-6*10^{10} M_sun). We find that at the median redshift of z=2.07, the median metallicity of the lensed sample is 0.35 dex lower than the local SDSS star-forming galaxies and 0.18 dex lower than the z ~ 0.8 DEEP2 galaxies. We also present the z ~ 2 MZ relation using 19 lensed galaxies. A more rapid evolution is seen between z ~ 1->3 than z ~ 0 -> 1 for the high-mass galaxies (10^{9.5-11} M_sun), with almost twice as much enrichment between z ~ 1 -> 3 than between z ~ 1 -> 0. We compare this evolution with the most recent cosmological hydrodynamic simulations with momentum driven winds. We find that the model metallicity is consistent with the observed metallicity within the observational error for the low mass bins. However, for higher masses, the model over-predicts the metallicity at all redshifts. The over-prediction is most significant in the highest mass bin of 10^{10-11} M_sun.