J784S4XEVM: Run Own Program on Evaluation Board

Hey,

For running a model on EVM, you just need artifacts, model and a script similar to onnxrt_ep.py available on edgeai-tidl-tools. This should work without any issue if version compatibility are followed. You can check more on edgeai-tidl-tools repo.

About your script.py which use some other packages like torch , if the packages are installed correctly then there should not be any issue using them. I tried pip3 install torch and used it on EVM, it works for me. Could you check if packages are installed correctly , like are simple things working like import torch etc ?

Hi thank you for your fast response. So I would have to manually pip install every module i am working with? I tried running singularity with a container and it said singularity command not found, and I m having difficulty installing. I also want to use some prebuilt wheels, how can i use those? 

is it possible I need a new image on the SD card?

Hi, I would suggest you to use pip install -r requirements.txt , you can also use prebuild wheels if they are compatible with EVM. I have never tried docker or singularity for TIDL on EVM, do you have any guide or instructions you are following for this setup? If you can provide more details, i can continue to help.

Usually I run under the edgeai_tidl_tools_x86_ubuntu_22 docker image. And i install all of these requirements under my requirements.txt 

line-profiler
wandb
imageio
lanelet2
shapely
omegaconf
hydra-core
pycountry
artifacts/pytorch3d-0.7.7-cp310-cp310-linux_x86_64.whl
https://download.pytorch.org/whl/cu118/torch-2.2.2%2Bcu118-cp310-cp310-linux_x86_64.whl#sha256=676b99efd763abd76cc4b3f70711ee2a0b85bdeafb656d4365740c807abbba69
https://download.pytorch.org/whl/cu118/torchvision-0.17.2%2Bcu118-cp310-cp310-linux_x86_64.whl#sha256=69ea7cba762b958a74e8accbb1f92d822b59497b36d3aecae759bdbab75f85d2
pytorch-lightning==1.9.0
sqlalchemy==2.0.4
sqlalchemy-file
comet_ml
loguru
apache-libcloud
psycopg2-binary
fasteners
invertedai
pyarrow
traitlets
transformers

this is what my run.sh script looks like to create the environment

docker run -w /home/root -it --rm --shm-size=4096m -v /shared:/shared --mount source=$(pwd),target=/home/root,type=bind edgeai_tidl_tools_x86_ubuntu_22 \
    bash -c "
        set -e

        if [ ! -d \"$VENV_DIR\" ]; then 
            echo ---------------------------
            echo SETTING UP PYTHON VENV
            echo ---------------------------

            apt update
            apt install -y python3.10-venv
            python3 -m venv $VENV_DIR
        fi

        if [ ! -d \"$ARTIFACTS_DIR\" ]; then
            echo artifacts/ folder not found. Exiting... 
            exit
        else
            source $VENV_DIR/bin/activate
            export SOC=$SOC
            source setup.sh

            rm -rf build lib bin
            mkdir build
            cd build/
            cmake ../examples/
            make
            cd ../
            
            pip install -r requirements.txt
        fi

        exec bash
    "

So, what exactly is the issue in this script ? Is 'pip install -r requirements.txt' not working on evm?

yes exactly it’s not working on evm

Could you share the output you are getting after running, pip install -r requirements.txt ? Also, if you have instructions to setup your docker style evm, i might be able to reproduce the issue at my end and help further.