Accelerate PyTorch with DeepSpeed to train large language models with Intel Habana Gaudi-based DL1 EC2 instances