Polymers are nonlinear materials showing phenomena such as stress relaxation, creep and stain-rate dependent yield. Constitutive models are frequently applied that incorporate rate-dependent mechanisms such as the Eyring process. To evaluate the parameters that define such mechanisms, experiments are conducted at specified strain rates. It is usually assumed that the strain rate in the rate dependent process is the same as, or simply related to, the applied macroscopic strain rate. However, as all polymers are inhomogeneous at some length scale, the macroscopic strain rate represents an averaging of the strain rate throughout the material. We analyse 3D representative volume elements (RVEs) representing semi-crystalline structures that consist of orthotropic elastic crystallites embedded in a viscoplastic amorphous polymer matrix. The matrix is modelled using a Maxwell-type approach consisting of an Eyring process in series with a hyperelastic neo-Hookean element. Solid crystallites are generated at random orientations within the matrix. Uniaxial deformations are applied to the RVEs and the overall stress-strain response is obtained. This is compared with the response of the amorphous polymer model for a range of strain rates. The material parameters correspond to a polylactic acid (PLA) polymer with 13% crystallinity at a temperature of 60C. The RVEs give realistic stress-strain responses for experiments on the PLA material at the initial stages of deformation up to ~ 4% macroscopic strain. Within this range, it appears that the crystal structure remains substantially intact. However, the strain-rate dependence of yield for the RVEs differs markedly from that of the viscoplastic matrix model. The relationship between the yield stress of the RVE and that of the matrix material is dependent on the macroscopic rate of strain. The modelling provides a means of deriving the underlying yield behaviour from macroscopic experiments.