We generate entanglement between two transmon qubits each coupled to a readout resonator and located on different chips via engineering a half-parity measurement. This measurement does not distinguish |01> from |10>, thus if started in an initial full twoqubit superposition it will project the qubits to an odd Bell state 50% of the time. By post-selecting on measurement outcomes corresponding to |00> or |11> we have generated Bell-states of 51% (40%) concurrence or 75% (70%) Bell-state fidelity keeping 25% (50%) of the data. With a repetition rate of 5KHz based on state initialization by T1 this corresponds to a 1KHz generation rate. We show that changing the cavity frequencies with tuning qubits and by driving with a second compensation pulse in a weakly coupled port we can reduce residual distinguishability in the odd subspace, making the protocol robust to fabrication imperfections. By shaping the pulse of the compensation drive we also generate an even-parity Bell state of similar concurrence. We found that thermal excitation results in an overestimate of concurrence in state tomography and show a simple method to correct for this by assuming mixed calibration points. We find good agreement of unconditioned measurement dynamics with theory using a reduced two-qubit only master equation. A 45% measurement efficiency was found using a stochastic master equation to model final state concurrence conditioned on the measurement result. This protocol could be used in modular 2D quantum computing architectures where entanglement is distributed with itinerant photons.