This repo contains the code we used to train a recurrent-depth model at scale on 4096 AMD GPUs on Frontier. All details on this model can be found in the tech report ...