TDA4VH-Q1: The issue of inference with configuration "inDataPadInTIDL"

xu guanghui

Part Number: TDA4VH-Q1

There are two models named "35414.tar" and "35447.tar".

We deployed the two models using configuration "inDataPadInTIDL: 1". However there are different issue with the two models.

The model named "35414.tar" will hang in tidl when inferring.

The model named "35447.tar" will get incorrect output.

We have also tried other models with the same configuration and they all work fine. Please help me check what the problem is.

35414.tar 35447.tar

over 1 year ago

0 Pratik Kedar over 1 year ago

TI__Mastermind 24041 points

Hi,

Which SDK version you are using ?

Please share the logs with debugTraceLevel set to 2 for both of the case.

0 xu guanghui over 1 year ago in reply to Pratik Kedar

Prodigy 20 points

Thank you for your reply.

The SDK version is 8.6.

The issue has been agreed upon with the TI local FAE team. Please reproduce and analyze it according to the attachment.

0 Pratik Kedar over 1 year ago in reply to xu guanghui

TI__Mastermind 24041 points

Hi,

Thanks for the reply.

As next process, we will need requested logs for further analysis as explained here.

Pratik Kedar said:
Please share the logs with debugTraceLevel set to 2 for both of the case.

Secondly,

Could you please share the observation on the same with latest 9.1 SDK, recent release has significant improvement wrt 8.6 this will null out possibility issue in lasted sdk and help to drive this correctly.

0 Yong Zhang over 1 year ago in reply to Pratik Kedar

TI__Genius 16631 points

Dear Guanghui.

as you mentioned, the two models doesn't work with "inDataPadInTIDL: 1". may I ask if they work well with opposite configuration, "inDataPadInTIDL: 0"?

BTW, would you please help share the update on your side on the suggestion from Kedar, "share the observation on the same with latest 9.1 SDK",?

thanks a lot!

yong

Processors

Processors forum

TDA4VH-Q1: The issue of inference with configuration "inDataPadInTIDL"