• State TI Thinks Resolved
  • Locked Locked
  • Replies 1 reply
  • Answers 1 answer
  • Subscribers 98 subscribers
  • Views 624 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

BEAGLE-3P-BBONE-AI: Deploy an AWS Sagemaker Model on BeagleBone AI

Lucas Forero
Prodigy 10 points

Part Number: BEAGLE-3P-BBONE-AI
Other Parts Discussed in Thread: AM5729

Hi,

We are currently trying to deploy an AWS Sagemaker object recognition model in the BeagleBone AI, using some of the reduction tools for embedded systems that they offer (the Sitara AM5729 is on the supported devices list). When we tested the model on the BeagleBone AI, we had poor performance (similar to the performance we get on the BeagleBone Black), seems like the AM5729 was working at minimum capacities.

If you know how to implement an AWS Sagemaker model on BeagleBone AI or you can help us configure the AM5729, ensuring maximum performance, it would be great.

Thanks a lot.

  • 0
    TI__Mastermind 40600 points

    Hi, we recently updated AWS Sagemaker to work with Subgraphs (non-supported layers run in ARM, the rest of a model layers run on TIDL HWA). Previously, it was more like a binary decision, if any layer was not supported by TIDL, then the whole model graph run on ARM.

    I am wondering if the models that you are testing were before we enabled Subgraphs in AWS Sagemaker. And if so, that could be an explanation on why you see a similar performance with BBAI and BBB boards..

    In any case, FYI, . We have released a patch and updated user’s guide to support this feature:

    Thank you,

    Paula