cuda-python==12.9.0 cupy-cuda12x numba numba-cuda[cu12] triton; platform_system != "Windows" torch --extra-index-url https://pypi.ngc.nvidia.com polygraphy colored numpy==1.26.4 onnx==1.18.0; platform_system == "Windows" --extra-index-url https://pypi.ngc.nvidia.com onnx-graphsurgeon pywin32; platform_system == "Windows" pyyaml==6.0.3 requests==2.32.4 tqdm==4.66.4