rcocorad
commented
3 months ago Update: I was able to get the train/eval scripts to run by creating a new conda environment from python 3.8, pip installing the requirements.txt, updating torch to 2.2.1+cu121, torchvision to 0.17.1+cu121, installing xir manually via the conda 3.5.0 channel, and isntalling pytorch nndct from script in the vitis-ai-3.5.0 repo.
It would be nice to get a docker image or conda environment that works out of the box.
I'm trying to build a yolov7 xmodel following the Readme and the jyupter noteback tutorial. I immediately get a string of errors then get stuck. I'm using docker vitis ait 3.5 gpu: Docker Image Version: 3.5.0.001-8108149 (GPU) Vitis AI Git Hash: 8108149 Build Date: 2024-02-26 WorkFlow: pytorch
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This pytorch-nndct 3.5.0 requires ninja, which is not installed. pytorch-nndct 3.5.0 requires scipy<=1.9.3, but you have scipy 1.10.1 which is incompatible.
ImportError: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.29' not found ( required by /opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.8/site-packages/pandas/_libs/window/aggregations.cpython-38-x86_64-linux-gnu.so)
OSError: libcublas.so.11: ELF load command address/offset not properly aligned
This seems to be a never-ending path of fixing environment build errors. Can anyone tell me what configuration is used to get the base tutorial working? It seems that since all the posts are 6+ months old maybe the rest of you are using vitis-ai-3.0? Not sure if anyone monitors this repo, so at the least this post identifies issues with vitis-ai-pytorch-gpu:3.5.0.001.