remove synchronize for pack_unpack_tensor_with_numpy

[Kaihui-intel]

Type of Change

feature or bug fix or documentation or validation or others
API changed or not

Description

detail description

Expected Behavior & Potential Risk

the expected behavior that triggered by this PR

How has this PR been tested?

how to reproduce the test (including hardware information)

Dependency Change?

any library dependency introduced or removed

[xin3he]

@Kaihui-intel let's wait till the Habana software 1.19.0 is merged back.

[thuang6]

could you give more background on this PR? why it is related to 1.19? is it expected to be cherry-picked to 1.19 release branch in INC fork?

[Kaihui-intel]

This PR is related to version 1.19 because it addresses a performance optimization related to unnecessary synchronization. The synchronize operation impacts the time spent on unpacking tensors on the HPU, and the unpack_tensor_with_numpy function, which only occurs on the CPU, does not require this synchronization. Therefore, the synchronize call is redundant and can be removed to improve efficiency.
As for the cherry-pick, it needs to be answered by @xin3he

[xin3he]

@thuang6 it's related with AutoRound PR and has conflict with it. so we should not merge this PR and wait util the AutoRound PR is cherry-picked back.
https://github.com/habana-internal/neural-compressor-fork/pull/25