Use VRS, we use CRS when there is no scale inefficiency which is mostly impossible. The answer of VRS will be equal to CRS if there is no scale inefficiency otherwise VRS may be different and best to estimate. In any case, VRS is necessary to calculate therefore it is better to go with it.

You may read the following citation for further learning:

Coelli, T. J., Rao, D. S. P., O'Donnell, C. J., & Battese, G. E. (2005). An introduction to efficiency and productivity analysis: Springer Science & Business Media.

You can use both CRS or VRS DEA Models to measure the efficiency score of DMUs and compare the results.

Charnes-Cooper- Rhodes or CCR model work on constant return scale whereas Banker-Charnes-Cooper model or BCC Model work on Variable return scale.

See the given book

Handbook on Data Envelopment Analysis by William W. Cooper, Lawrence M. Seiford, Joe Zhu

Variable returns to scale (VRS) is a frontier scale used in data envelopment analysis (DEA). It helps to estimate efficiencies whether an increase or decrease in input or outputs does not result in a proportional change in the outputs or inputs respectively (Cooper, Seiford, & Zhu, 2011). This method includes both increasing and decreasing returns to scale. Hence, VRS may have more realistic results in comparison to the CRS scale.