[Torch, CI] Update to PyTorch 1.7

[masahi]

A part of #6594. Updated frontend tests and the install script to install torch==1.7.0 torchvision==0.8.1

@tqchen so according to https://tvm.apache.org/docs/contribute/pull_request.html#ci-environment, the next step after this PR is to build and upload a new docker image, and update Jenkinsfile. Since I have a committership, I can do those steps, is that right? But I've just created an account with dockerhub (https://hub.docker.com/u/masahi129), so I don't think I have a privilege to push under tlcpack org).

I'm not familiar with docker workflows, but I want to learn how to update CI.

[tqchen]

@masahi you should now have access

[masahi]

@tqchen I've uploaded ci-gpu:v0.72 to https://hub.docker.com/repository/docker/tlcpack/ci-gpu, and updated Jenkinsfile to point to the new image, but it seems jenkins is running the older v0.71 image. What am I missing?

[leandron]

Had a quick look. A couple minor comments.

[tqchen]

@masahi We don't use images from Jenkinsfile in PRs due to security concerns. the Jenkinsfile have to be pushed into ci-docker-staging(you can push directly to that branch) for stage testing. Then the results will be available here https://ci.tlcpack.ai/job/tvm/job/ci-docker-staging/

After the stage testing passes, we then send the PR to the mainline to update the image. Note that you may want to create temporary tags like v0.72-t0 in cases where we want to build the docker image multiple times, this usually happens when an error occurs during the upgrade

[masahi]

Tests on the staging branch https://ci.tlcpack.ai/job/tvm/job/ci-docker-staging/34/ have passed, but this PR is still running on the older image. I'll close this one and try open another PR.