Add support for MRPC dataset with unit tests

[vcm2114]

Summary

• Added support for MRPC dataset
• Added mocked unit tests for MRPC dataset

Test

• pytest test/datasets/test_mrpc.py

Context

See #1710

[parmeet]

LGTM! Please fix lint issues before merging :)

[Nayef211]

@ejguan @NivekT it seems like some of the CI builds are failing at the step where we try to install torchdata from source (i.e. here). Do you know if there's been any recent changes that would cause this?

[ejguan]

@ejguan @NivekT it seems like some of the CI builds are failing at the step where we try to install torchdata from source (i.e. here). Do you know if there's been any recent changes that would cause this?

This seems weird to me. TorchData is required PyTorch to run some codegen. Based on the script (https://github.com/pytorch/text/blob/main/.circleci/unittest/linux/scripts/install.sh), it seems PyTorch is installed before TorchData. Not sure why this happens.
Please correct me if I am wrong. It seems the CI shouldn't use source to install TorchData

text/.circleci/unittest/linux/scripts/install.sh

Lines 32 to 33 in ab76a04

	printf "Installing torchdata nightly\n"
	pip install --pre torchdata --extra-index-url https://download.pytorch.org/whl/nightly/cpu

.

[NivekT]

@ejguan @NivekT it seems like some of the CI builds are failing at the step where we try to install torchdata from source (i.e. here). Do you know if there's been any recent changes that would cause this?

This seems weird to me. TorchData is required PyTorch to run some codegen. Based on the script (https://github.com/pytorch/text/blob/main/.circleci/unittest/linux/scripts/install.sh), it seems PyTorch is installed before TorchData. Not sure why this happens. Please correct me if I am wrong. It seems the CI shouldn't use source to install TorchData

text/.circleci/unittest/linux/scripts/install.sh

Lines 32 to 33 in ab76a04

	printf "Installing torchdata nightly\n"
	pip install --pre torchdata --extra-index-url https://download.pytorch.org/whl/nightly/cpu

Maybe switching to install nightly with conda will help?

[Nayef211]

Please correct me if I am wrong. It seems the CI shouldn't use source to install TorchData

@ejguan my mistake I guess we are relying on torchdata nightlies from pypi in our unit tests setup and not installing directly from source.

Maybe switching to install nightly with conda will help?

@NivekT do you know what the correct command would be to install the nightly from conda?

text/.circleci/unittest/linux/scripts/install.sh

Lines 32 to 33 in ab76a04

	printf "Installing torchdata nightly\n"
	pip install --pre torchdata --extra-index-url https://download.pytorch.org/whl/nightly/cpu

[Nayef211]

@NivekT do you know what the correct command would be to install the nightly from conda?

Nvm I just found the following link https://anaconda.org/pytorch-nightly/torchdata. I assume it would be as follows

conda install -c pytorch-nightly torchdata

[ejguan]

I think this command should be fine:
pip install --pre torchdata --extra-index-url https://download.pytorch.org/whl/nightly/cpu
I just tested it on my local machine.

[Nayef211]

LGTM. Delete print lines before merging