GPT-OSS with Flash Attention and Memory-Efficient Attention via PyTorch-Native SDPA from time import time import warnings import torch import transformers