Riva Question and Answering Model for ARM64 and Xavier NX

[Mr.razzaghi]

Please provide the following information when requesting support.

Hardware - GPU Xavier NX
Hardware - CPU 6 Core
Operating System 18.04
Riva Version 2.0.0 and 2.2.0
TLT Version (if relevant)
How to reproduce the issue ?
I want to use Riva Question Answering on My Xavier NX as Embedded .
When i add rmir_nlp_question_answering_bert_base to my nlp models TRT start to ready model for my device and i get below error:
[TensorRT] ERROR: Tactic Device request: 1881MB Available: 1536MB. Device memory is insufficient to use tactic.
As I see nvidia team ready some models for Xavier NX I want to know when QA model will be release for Xavier NX or it is so far please guide me to fix this issue

My Config file is:
config.sh (11.3 KB)

Error Screenshot:

[rvinobha]

Hi @Mr.razzaghi

Thanks for your interest in Riva

Apologies you are facing the issue

Thanks for sharing the config,

Since we want to use nlp based model only and memory related error, can we try to change the config.sh line no 15,16 and 17 like below

# Enable or Disable Riva Services
service_enabled_asr=false
service_enabled_nlp=true
service_enabled_tts=false

and let us know if the issue/error still persists (riva_clean.sh before trying out the above), if possible i request to kindly share the output of riva_init.sh and riva_start.sh

Thanks