JVET-AJ0148 proposed Transformer-based reference frame synthesis for VVC inter coding, named TRFS. This contribution reports the test results of TRFS trained on the BVI-DVC dataset in RA configuration. Following the training strategy proposed in JVET-AJ0148, a new training set is generated on the BVI-DVC dataset using VTM-11.0_NNVC-9.0. Then, TRFS is retrained on the new training set in RA configuration. For tests, the retrained TRFS model is integrated in the latest VTM-11.0_NNVC-11.0 reference software. Moreover, this contribution further reduces the computational complexity by replacing regular convolutions with depthwise separable convolutions. Compared to VTM-11.0_NNVC-11.0, the BD-rate performance of retrained TRFS is as follows: