Help Request: PyTorch / ONNX Compatibility on NVIDIA Clara AGX
Hello,
I’m working with an NVIDIA Clara AGX system with the following specifications:
-
OS: Ubuntu 20.04.6 LTS (
focal) -
nvidia-l4t-core: Version 34.1.2-20221208104417
From what I understand, this setup is not officially tied to Jetson JetPack releases, so I don’t have a clear reference for software compatibility. Most online compatibility charts for PyTorch / ONNX are listed against JetPack versions, not directly for Clara AGX.
My questions:
-
How can I determine which versions of PyTorch and ONNX Runtime are compatible with Clara AGX, given my environment?
-
Is there a way to update the system so that compatibility information (normally tied to JetPack versions) applies here?
-
If only pytorch 1.x is compatible with the Clara, is it better to keep my pytorch 2.3 that I built from source or use the latest compatible pytorch (i.e. 1.x)? My goal is to make a diffusion model run as fast as possible
What I’ve tried so far:
-
I successfully built PyTorch from source:
-
Version:
2.3.0a0+git63d5e92 -
CUDA arch:
sm_75(setTORCH_CUDA_ARCH_LIST="7.5;7.5+PTX") -
Disabled:
USE_FLASH_ATTENTION,USE_MEM_EFF_ATTENTION,USE_MKLDNN,USE_NCCL
-