PT2E QAT Move model to eval before convert

[leslie-fang-intel]

Description

As pytorch/pytorch#130395 introduced a robust DCE pass, it exposed an issue with the QAT Conv-BN folding path when we write testcase and usage. Since we have fixed this in the PyTorch main branch in pytorch/pytorch#130598, we should update the tutorial accordingly.

Checklist

• The issue that is being fixed is referred in the description (see above "Fixes #ISSUE_NUMBER")
• Only one issue is addressed in this pull request
• Labels from the issue that this PR is fixing are added to this pull request
• No unnecessary issues are included into this pull request.

cc @jerryzh168 @z-a-f @vkuzo @jgong5 @mingfeima @XiaobingSuper @sanchitintel @ashokei @jingxu10 @ZailiWang @ZhaoqiongZ @Xia-Weiwen @sekahler2 @CaoE @zhuhaozhe @Valentine233

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/tutorials/2977

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit c2db6b6 with merge base 2f2db74 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[leslie-fang-intel]

Hi @svekars, could you please help to take a look of this PR?

[leslie-fang-intel]

Thanks. Is it ok to land now?

[leslie-fang-intel]

Let's wait for the discussion of the ordering between move_exported_model_to_eval and convert_pt2e. cc @andrewor14 @yushangdi

[leslie-fang-intel]

Close it as @andrewor14 has fixed it. We don't need to adjust the order.