feat: add minimal strict selected-logprob provider hook

[inaniloquentee]

Summary

Add the smallest Vime-side integration point needed to consume the RL-Kernel TP-aware selected-logprob provider for the Qwen3-8B TP=2, CP=2 post-training path.

Scope

• Add a structural provider boundary for selected logprob computation.
• Keep CP token-row layout, response extraction, and PPO/GRPO loss composition in Vime.
• Add strict provider validation and reject unsupported RL-Kernel top_p replay (top_p must be 1.0 for this validated path).
• Allow the selected-logprob path to receive BF16/FP16/FP32 logits while preserving the existing value-output dtype contract.
• Add a Qwen3-8B TP=2, CP=2, PP=1 launch script with no RL-Kernel-specific Attention/FFN wrapper copied into Vime.

The provider implementation, contract, provenance, runtime evidence, and Qwen3 validation example live in RL-Kernel PR #289.

Validation

Check	Result
Megatron argument/provider tests	21 passed
Python compileall	passed
git diff --check	passed
Real Qwen3-8B multi-GPU run	pending matching GPU environment

The launch script is intentionally fail-closed: strict mode requires the provider and rollout-top-p=1.0; it does not claim a successful run from configuration alone.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: e5db969c-6fa4-4f98-8189-6c4d773b83a7

---

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.}