Add benchmarks for CvT/BEVT/MoE/Oscar

[mindest]

Adding Benchmark cases for four models:

• CvT (Convolution vision Transformers)
• BEVT (BERT Pretraining of Video Transformers)
• MoE (Mixture of Experts)
• Oscar (Object-Semantics Aligned Pre-training for Vision-and-Language Tasks)

This is a work in progress.

Checklist:

• [x] Add benchmark case for CvT
• [x] Add benchmark case for BEVT
  • [x] Add draft benchmark for BEVT
  • [x] Add data downloading for BEVT model
• [x] Add benchmark case for MoE
• [x] Add benchmark case for Oscar
• [x] Add Dockerfiles for benchmark environment prep
• [x] Add data access for Oscar

[facebook-github-bot]

Hi @mindest!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@fb.com. Thanks!

[samiwilf]

@mindest CvT, MoE, and Oscar look good to me, but I hit an IndexError in Bevt. I'm not using the docker container for the MI250 runs as that's discouraged to do on the servers that run the MI250 GPUs I'm using. Here is the log of the IndexError: bevt-apex-error.log

[zstreet87]

@mindest CvT, MoE, and Oscar look good to me, but I hit an IndexError in Bevt. I'm not using the docker container for the MI250 runs as that's discouraged to do on the servers that run the MI250 GPUs I'm using. Here is the log of the IndexError: bevt-apex-error.log

I believe this is what @mindest is cherry-picking for in the dockerfile. see https://github.com/NVIDIA/apex/pull/1282

[mindest]

@mindest CvT, MoE, and Oscar look good to me, but I hit an IndexError in Bevt. I'm not using the docker container for the MI250 runs as that's discouraged to do on the servers that run the MI250 GPUs I'm using. Here is the log of the IndexError: bevt-apex-error.log

Hi @samiwilf, as @zstreet87 mentioned the fix NVIDIA/apex#1282 is needed for apex. If you are able to docker pull the prebuilt image mindest/rocm5.2_ubuntu20.04_py3.7_pytorch_1.11.0:bevt it will save you time.

[samiwilf]

@mindest, I wish I could use docker. There are internal Meta reasons docker isn't allowed on the MI250 computer I'm using.

[zstreet87]

@mindest, I wish I could use docker. There are internal Meta reasons docker isn't allowed on the MI250 computer I'm using.

@samiwilf Try something like this (replacing wherever the utils.py script is) sed -i 's/next_functions\[1\]/next_functions\[0\]/g' /opt/conda/lib/python/site-packages/apex/amp/utils.py

[mindest]

@mindest, I wish I could use docker. There are internal Meta reasons docker isn't allowed on the MI250 computer I'm using.

@samiwilf Try something like this (replacing wherever the utils.py script is) sed -i 's/next_functions\[1\]/next_functions\[0\]/g' /opt/conda/lib/python/site-packages/apex/amp/utils.py

This should work, or alternatively just disable/comment out check on these lines as workaround.

[facebook-github-bot]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!

[facebook-github-bot]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!