Text this: Distributed B-SDLM: accelerating the training convergence of deep neural networks through parallelism