Open oak-barry opened 2 weeks ago
hi,
if you choose using cuda12.1, and you install onnxruntime_gpu with pip install onnxruntime_gpu==1.17.0 directly, in the runtime, it will throw
UserWarning: Specified provider 'TensorrtExecutionProvider' is not in available provider names.Available providers: 'AzureExecutionProvider, CPUExecutionProvider'
this is because the default onnxruntime_gpu match to cuda 11.8, not 12.1, according to https://onnxruntime.ai/docs/install/
and this warning will cause the speed slow down.
for cuda 12.1, as the doc I refered, we need to run it instead: pip install onnxruntime-gpu --extra-index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/simple/
this is also the reason I do the uninstall then re-install step in https://github.com/AiuniAI/Unique3D/issues/15 and bat I commited
for the repo owners, their orignal env is cuda 11.8, which is ok to run pip install onnxruntime_gpu directly
remove requirements-detail.txt and merge it into requirements.txt update README.md, configure installation on linux