[run_parallel] generate correct output filename

the filename does not integrate the model_size --> impossible to run inference with different model size without destroying data!