Training and inference estimates derive operation counts from tensor shapes and the number of passes. The measure does not capture memory traffic, communication, sparsity, or hardware utilization, so equal FLOPs need not mean equal runtime.
FLOPs count floating-point arithmetic operations and are used to estimate the compute required by a model or workload.
Training and inference estimates derive operation counts from tensor shapes and the number of passes. The measure does not capture memory traffic, communication, sparsity, or hardware utilization, so equal FLOPs need not mean equal runtime.