# the fast CUDA backend of qev.py (qev_fast/): pip install -r requirements.txt -r requirements-cuda.txt
flash-linear-attention==0.5.2