[CMake error] Building Triton on arm64 machine using build.py

[crane-sapia]

Settings

I am trying to build a Triton server from build.py with r23.12 on an AWS EC2 instance with

• platform: Amazon Linux (Inferred).
• chip: AWS Inferentia (arm-based architecture)
• cmake version 3.22.2
• python version 3.8

The following is the script that I use to build the server:

python build.py \
--enable-logging \
--enable-stats \
--enable-tracing \
--enable-metrics \
--no-container-interactive \
--endpoint=http \
--endpoint=grpc \
--endpoint=sagemaker \
--repo-tag=common:r23.12 \
--repo-tag=core:r23.12 \
--repo-tag=backend:r23.12 \
--repo-tag=thirdparty:r23.12 \
--backend=ensemble:r23.12 \
--backend=identity:r23.12 \
--backend=repeat:r23.12 \
--backend=square:r23.12 \
--backend=onnxruntime:r23.12 \
--backend=pytorch:r23.12 \
--backend=python:r23.12 \
--repoagent=checksum:r23.12

Error Description

I am getting the error shown below:

Can anyone help me solve this issue?

[yinggeh]

Hi @crane-sapia. Can you rerun with --image=gpu-base,nvcr.io/nvidia/tritonserver:<xx.yy>-py3-min flag?
You can read more about CPU-only build here.

[crane-sapia]

Hi @yinggeh thx for your response. The build.py file now runs successfully. However, while I was checking the docker images, I saw the following list of images. May I know which one that I can use as a base image for my sagemaker program?

[yinggeh]

Hi @crane-sapia. You should use the second image tritonserver and you can see it was just created.
nvcr.io/nvidia/tritonserver is downloaded from public release. Please check out Run on CPU-Only System for more info.

[crane-sapia]

Hi @yinggeh Thank you for your help. I will give it a try and see if it works as the docker image provided by aws as shown below