reasoning model evaluation mmlu gpqa

[ruchaa-apte]

What does this PR do ?

This PR adds Part 2 of the tutorial, which covers the evaluation phase.

Usage

python prepare_dataset.py --datasets [mmlu, gpqa, gpqa_diamond, all]
python deploy_nemo_model.py --checkpoint_path <checkpoint_path> --dataset <dataset> --output_prefix <output_prefix>
python evaluate_model.py --input_file <input_file> --output_file <output_file> --model_name <model_name>

Before your PR is "Ready for review"

Pre checks:

• [Y] Make sure you read and followed Contributor guidelines
• [N] Did you write any new necessary tests?
• [Y] Did you add or update any necessary documentation?
• [N] Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• [Y] New Feature
• [N] Bugfix
• [N] Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

[ruchaa-apte]

cc - @Maghoumi , @aasthajh

[Maghoumi]

I noticed this loop takes a very long time to go through, and GPU utilization is pretty low.

Is this a legit triton server, with support for multiple parallel queries?

If so, we could easily parallelize this with multiple threads and significantly improve both runtime and GPU utilization.

[ruchaa-apte]

Hey @Maghoumi I checked with NeMo eng team and they mentioned that this cannot be parallelized, but there is another PR they are trying to raise which addresses this concern :(