Skip to content

[AYCI] Arm Tvm converter, inference lambda 코드 이상, tvm 환경 이상 체크 #29

@workdd

Description

@workdd

현재 step function 시스템 내에서 성능 측정을 해보는 중입니다.
그 결과 예전에 arm tvm에서 추론 성능을 측정해보았을 때보다 성능이 정상적이지 않은 수준으로 안좋게 나오는 상황입니다.

<style type="text/css"></style>

inception_v3 torch onnx tvm torch onnx tvm
convert time   8.56 89.5   9.21 565.59
inference lambda time 11.43 2.46 13.53 4.61 4.57 66.65

inception_v3를 대표적으로 보았을 때, 565초의 convert time과 1분 이상의 추론 시간이 소요되는 것은 세팅에 문제가 있다고 판단이 됩니다. 따라서 이를 한번 체크해보아야 할 듯 합니다.

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions