SK-TDA4VM: Testing fail with pre-compiled model in edgeai_tools

Hi,

Could you share some more details about your tests? What SDK and which version are you using? How are you testing the pre-built models?

Thank you,

Fabiana

Hello Fabiana.

Following the guidelines, I have set up edgeai_tidl_tools in the Docker based x86_PC.
(github.com/.../README.md)

I chose to use the master branch after seeing a thread indicating issues with the previous versions of edgeai_tidl_tools.
(e2e.ti.com/.../4880253

After that, build the examples and run the script for testing.


When I use the prebuilt image for SK-TDA4VM in PSDK 09.00.01.03, I observe the same issue on the board.


for more detail 
create docker with below command. and install packages manually

- docker run  command : docker run -it --publish 8888:8888 --gpus all --shm-size=4096m --mount source=$(pwd),target=/home/root,type=bind --name edgeai-tidl-tools_${USER} x86_ubuntu_22

Because of the Thanksgiving holiday in the U.S., TI E2E design support forum responses may be delayed the week of Nov. 20.

Thank you for your patience.

Hi,

Can you please try out model compilation on latest 9.1 sdk ? (We have added few fixes in this release, this might solves the problem)

Secondly if you are facing the issue with setting up the docker, you can check out our collab notebooks here : 

1. https://github.com/kumardesappan/colab-notebooks/blob/main/ti_edgeai_tidl_tools_9.0.ipynb (AM68PA/TDA4VM)
2. https://github.com/kumardesappan/colab-notebooks/blob/main/ti_edgeai_tidl_tools_9.0_am68a.ipynb (AM68A/TDA4VL)
3. https://github.com/kumardesappan/colab-notebooks/blob/main/ti_edgeai_tidl_tools_9.0_am69a.ipynb (AM69A/TDA4VH)

hi Pratik.

I tried to run the latest version of edgeai-tidl-tools (tag: 09_01_00_01), but I got the same report as the previous version.

- log : run_python_examples.sh

/cfs-file/__key/communityserver-coponents-multipleuploadfilemanager/86d232b0_2D00_b496_2D00_4ad6_2D00_a622_2D00_080b02e26e30-379840-complete/log_2D00_run_5F00_python_5F00_examples.txt

- log : gen_test_report.py

/cfs-file/__key/communityserver-discussions-components-files/791/log_2D00_gen_5F00_test_5F00_report.txt

- report : 

/cfs-file/__key/communityserver-discussions-components-files/791/test_5F00_report_5F00_pc_5F00_am68a.csv

Hi,

As I see, from the generated csv file.

I can classify this issue into 2 categories.

1. Output miss match

2. Output file missing 

Moreover, one observation could be few model which are failing in python flow are working in cpp flow, and vice versa.

May I know which model you are working closely(on priority) with and that is failing here ?

Can you try to run the same using edgeai tidl tools 9.1 separately and report your findings here (Am suspecting there could be something might be going wrong through automated test script) ? We can dive deeper for the particular model where you are able to reproduce the issue standalone with tidl tools.