LLM post training RCP logs and finalized description

[al-rigazzi]

Summary

Finalizes the description and reproducibility instructions for the Qwen 3.5 397B GRPO post-training benchmark.

Included

• Adds RCP logs for six replicas at each of GBS 256, 512, and 1024.
• Updates the NeMo-RL submodule to the qualified implementation at 3fca04c9b.
• Expands the benchmark description, system-independent setup instructions, configuration details, and timing methodology.
• Documents the exact model, datasets, Gym/OpenHands prompts, validation metric, and rollout-completion requirements used for the RCPs.
• Adds helpers for downloading the checkpoint and datasets and verifying dataset identity.
• Adds detailed instructions for building the container and launching reproducible RCP runs.

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[ShriyaRishab]

Can you please rename the benchmark folder to match the agreed upon name?

[al-rigazzi]

Can you please rename the benchmark folder to match the agreed upon name?

Totally an oversight from me.

@jepio could you do this? I don't have easy access to code for one week.

[ShriyaRishab]

I can make the change!
But you need to sign the CLA for us to be able to merge your MR.

[al-rigazzi]

I can make the change!

But you need to sign the CLA for us to be able to merge your MR.

I have enrolled into MLCommons and I believe I was added to the CLA. I haven't received a separate request for it.

[al-rigazzi]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

@ShriyaRishab seems to have worked