That change checks out then. They didn't see much need for FP16 outside of that so no longer run it at double FP32 rate outside of tensor cores (unless I'm mix

That change checks out then. They didn't see much need for FP16 outside of that so no longer run it at double FP32 rate outside of tensor cores (unless I'm mixing that up with AMD). Other forms of sparsity are heavily used at training time now, like block compression in Deepseek.