This work presents a machine learning approach for learning elastoplasticity directly from stress-strain data. Data-driven plasticity modeling is challenging due to the non-smooth transitions induced by the yield criterion and the complex nature of multi-dimensional yield surfaces. To address these difficulties, we propose a learning framework which bridges traditional constitutive modeling of elastoplasticity with learnable plastic yield surfaces. We build upon our previous work by accounting for learnable hardening mechanisms. From the machine-learning perspective, the proposed approach falls into the class of implicit layers and is implemented in a fully differentiable material modeling framework based on JAX. The proposed approach exhibits strong generalization capabilities due to its embedded structure, while requiring a moderate number of training parameters. Good performance with limited data and in the presence of noise is also observed.