Add GRPO fine-tuning example with README

[Fiona-Waters]

Summary

• Add comprehensive README for GRPO (Group Relative Policy Optimization) fine-tuning, following the pattern established by existing SFT, OSFT, and LoRA examples
• Add GRPO to the parent fine-tuning examples overview
• Example notebook (grpo_lora-kubeflow-trainjob.ipynb) demonstrates single-GPU GRPO training via Kubeflow SDK and Training Hub's ART backend

What's included

• examples/fine-tuning/grpo/README.md — algorithm overview, hardware requirements, workbench setup guide, GRPO-specific considerations (dshm volume, gpu_memory_utilization)
• examples/fine-tuning/grpo/grpo_lora-kubeflow-trainjob.ipynb — example notebook: TrainJob submission, parameter configuration, dataset format documentation, metrics inspection, reward curve plotting
• examples/fine-tuning/README.md — updated to list GRPO in the algorithm list and Distributed "Learn more" section

Dependencies

• Kubeflow SDK: LORA_GRPO algorithm support feat: Add lora_grpo implementation to training hub algorithms opendatahub-io/kubeflow-sdk#111
• Training Hub: 0.8.1 release.
• Universal Training Image with openpipe-art and updated training-hub

TODO

• Update %pip install to point to midstream once the SDK PR is merged, then remove once the workbench (universal) image includes the updated SDK
• Validate notebook on fresh workbench with final image

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Enterprise

Run ID: ae51e0bd-8a8b-46e9-8677-056264d41472

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[briangallagher]

@Fiona-Waters I would suggest interactive and distributed modes to be consistent with SFT, OSFT and Lora.

Is it possible to add a "Test the Trained Model" section in the notebook similar to the other algorithm notebooks.

Might be easier to add the MlFlow Interactive part as in this PR now, rather than have to open a follow on PR. Just for interactive example.

[Fiona-Waters]

@Fiona-Waters I would suggest interactive and distributed modes to be consistent with SFT, OSFT and Lora.

Is it possible to add a "Test the Trained Model" section in the notebook similar to the other algorithm notebooks.

Might be easier to add the MlFlow Interactive part as in this PR now, rather than have to open a follow on PR. Just for interactive example.

Thanks @briangallagher I have:

• added an interactive notebook
• Updated the readme to relect this
• Added a test the trained model section
• did not add MLflow as art does not support it.
  Please re-review.