• State Resolved
  • Locked Locked
  • Replies 3 replies
  • Answers 1 answer
  • Subscribers 100 subscribers
  • Views 206 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

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.

TDA4VM: Neural Network model Inference Memory management

Aneesh Peram
Expert 1255 points
Part Number: TDA4VM

Hi Team, 

I am trying to understand, how memory will be managed during the Neural Network model Inference on DSP.

Question1: Will the two .bin files(net.bin and io.bin, which are required for model inference) be loaded entirely into a ram from the SD card.

Question2: For example The size of two bin files is 5Mb then, Will the 5Mb be loaded entirely at application initialization and released at deinitialization (OR) It will be loaded and released only at inference time.

Could you please help me to understand the answers to these two questions?

Thanks and Regards,

Aneesh

  • 0
    TI__Guru 56820 points

    Hi Aneesh,

    Aneesh Peram said:
    Question1: Will the two .bin files(net.bin and io.bin, which are required for model inference) be loaded entirely into a ram from the SD card.

    Yes.

    Aneesh Peram said:
    Question2: For example The size of two bin files is 5Mb then, Will the 5Mb be loaded entirely at application initialization and released at deinitialization (OR) It will be loaded and released only at inference time.

    Yes its expected to be this way.

    Regards,

    Anshu

  • 0 in reply to Anshu Jain
    Expert 1255 points

    Hi Anshu,

    Thank you for the information.

    I understood the answer to the first question. But I did not understand the second question. Can you please elaborate on the second question answer?

    Thanks and Regards,

    Aneesh

  • +1 in reply to Aneesh Peram
    TI__Guru 56820 points

    Hi Aneesh,
        What i meant is model are expected to be loaded once during application initialization and are not expected to be loaded again for each frames inference. 


    Regards,

    Anshu