1 episode
Tensor, pipeline, or offload? What actually happens when you run one model across several GPUs — including mismatched ones.
