[luci] Remove Add_STR tests

[jinevening]

This removes invalid Add_STR tests.

ONE-DCO-1.0-Signed-off-by: Hyukjin Jeong [email protected]

---

Related to: https://github.com/Samsung/ONE/pull/13750/files#r1730675966
Draft PR: #13750

[seanshpark]

As the comment says

# This is test for import/export of STRING tensortype
# interpreter or runtime may fail as Add won't support this
``
this is just for STRING tensortype, happen to be using `Add` Op...
we should not remove this.

what test is failing?

[jinevening]

PTAL https://github.com/Samsung/ONE/pull/13750/files#r1730675966. STRING inputs are not valid for ADD. Those recipes should have CUSTOM operators.

[seanshpark]

... Those recipes should have CUSTOM operators.

I've read the comment but I don't know which test fails.
Asking again, which test fail?

[jinevening]

what test is failing?

Circle files for Add_STR are not created under common-artifacts after ADD's shape inference logic is updated.

[jinevening]

It seems that circle2circle is a problem. Add_STR_000.opt.circle is not created.

[seanshpark]

under common-artifacts after ADD's shape inference logic is updated.

can you add to to exclude.lst ?

[seanshpark]

tcgenerate(Add_STR_000) # STRING is not supported
tcgenerate(Add_STR_001) # STRING is not supported

um.. already added...

[jinevening]

I think tcgenerate is not enough. May need to add optimize(Add_STR_000) and optimize(Add_STR_001).

[seanshpark]

um... sorry for wrong questions. I was confused with the draft and changes of this PR.

[seanshpark]

LGTM

[seanshpark]

Anyway recipe itself should not be removed for tflchef.

[jinevening]

Anyway recipe itself should not be removed for tflchef.

I see. With #13754, it will be possible to leave the recipes. But, those recipes generate invalid tflite, so I think it would be better to remove them someday.

[seanshpark]

so I think it would be better to remove them someday.

Yes, they are invalid files but they exist for test creating STRING tensors, nothing more.