[quant] Backend string for the quantized types

[z-a-f]

Stack from ghstack:

• [quant] Backend string for the quantized types #49965 [quant] Backend string for the quantized types

Without this checking the type of the quantized tensor using type would throw an error.

After this PR running the type(qx), where qx is a quantized tensor would show something like torch.quantized.QUInt8.

Test Plan: Not needed -- this is just a string description for the quantized tensors

Differential Revision: D25731594

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit c42822e (more details on the Dr. CI page):

---

✅ None of the CI failures appear to be your fault 💚

---

• 1/1 broken upstream at merge base 97c17b4 on Dec 29 from 2:04pm to 6:33pm

---

1 job timed out:

• pytorch_linux_bionic_py3_8_gcc9_coverage_test1

---

🚧 1 fixed upstream failure:

These were probably caused by upstream breakages that were already fixed.

Please rebase on the viable/strict branch (expand for instructions)

If your commit is older than viable/strict, run these commands:

git fetch https://github.com/pytorch/pytorch viable/strict
git rebase FETCH_HEAD

Check out the recency history of this "viable master" tracking branch.

• pytorch_linux_bionic_py3_8_gcc9_coverage_test1 on Dec 29 from 2:04pm to 6:33pm (891759f - 4c5a4db)
  • 🔁 rerun

---

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions to the (internal) Dr. CI Users group.

This comment has been revised 5 times.

[jerryzh168]

currently we put qint8, quint8 and qint32 in torch namespace, do we need to change that?

[ezyang]

Do you need QuantizedCUDA too?

[z-a-f]

currently we put qint8, quint8 and qint32 in torch namespace, do we need to change that?

When we call x.type() it shows torch.IntTensor, while x.cuda().type() would show torch.cuda.IntTensor. I think the whole consensus is to have a backend prefix between torch and the tensor type. In fact, I think we should follow the pattern and show something like torch.quantized.UInt8Tensor instead of torch.quantized.QUInt8. @jerryzh168 What do you think?

[z-a-f]

Do you need QuantizedCUDA too?

How would the string look like? Something like torch.cuda.quantized or torch.quantized.cuda?

[ezyang]

IDK, but the answer to that question is wherever you actually put the Python modules for it.

[jerryzh168]

currently we put qint8, quint8 and qint32 in torch namespace, do we need to change that?

When we call x.type() it shows torch.IntTensor, while x.cuda().type() would show torch.cuda.IntTensor. I think the whole consensus is to have a backend prefix between torch and the tensor type. In fact, I think we should follow the pattern and show something like torch.quantized.UInt8Tensor instead of torch.quantized.QUInt8. @jerryzh168 What do you think?

sounds good. feel free to add the change to align quantized types with other types

[facebook-github-bot]

@z-a-f merged this pull request in 7ce8f7e.