docs: map vime architecture hook points

[inaniloquentee]

Summary

• Add a WS5 design note mapping Vime architecture and RL-Kernel hook points.
• Document the rollout-side path through RolloutManager, vLLM/vllm-router, Sample creation, and train-data conversion.
• Document the training-side path for Megatron logprob, loss, and reduction code.
• Identify candidate hook points for a minimal RL-Kernel PoC and capture WS3 handoff questions.

Scope

This is a documentation-only PR.

It does not inject RL-Kernel code, change Vime/slime/vLLM/Megatron source, claim numerical alignment, or add benchmark results.

Closes #118
Part of #83

Summary by CodeRabbit

• Documentation
  • Added comprehensive architecture documentation covering Vime's end-to-end training and rollout flow, including integration points, key modules, and extension capabilities for system customization and integration.

[coderabbitai]

📝 Walkthrough

Walkthrough

Adds docs/design/vime-architecture-and-hook-points.md, a new design document that maps Vime's rollout and training architecture, enumerates key source modules, traces token/logprob data flow, inventories RL-Kernel hook points with tensor contracts and stability notes, and records a recommended PoC path along with WS3 handoff findings and open questions.

Changes

Vime Architecture & Hook-Point Documentation

Layer / File(s)	Summary
Vime architecture and hook-point design document docs/design/vime-architecture-and-hook-points.md	New 243-line design doc enumerating Vime source modules, rollout-side and training-side token/logprob data flow, hook-point matrix with required tensor contracts and risk/stability notes, recommended PoC integration path, WS3 handoff findings, open questions, and explicit non-goals.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~5 minutes

Possibly related issues

• [WS5] RL-Kernel to vime integration design doc #119: This PR produces the Vime hook-point matrix and architecture map that issue #119 explicitly depends on to author its RL-Kernel-to-Vime integration design document.
• [WS5] Minimal vime PoC with one RL-Kernel operator #120: This PR documents the hook-points and recommends a PoC integration path that issue #120 targets for executing a minimal PoC by wiring an RL-Kernel operator into one of those hook-points.

Poem

🐇 A map of the hooks, a chart of the flow,
Where logprobs are born and tensors must go,
I traced every module with twitching delight,
And wrote down the contracts by soft moonlight.
The PoC path is clear — now let's code it right! 🗺️

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The pull request title 'docs: map vime architecture hook points' is directly related to and accurately summarizes the main change, which is adding documentation that maps Vime's architecture to RL-Kernel hook points.
Linked Issues check	✅ Passed	The pull request fulfills all primary coding-related objectives from issue #118: providing architecture overview, rollout-side and training-side execution paths, hook-point matrix, and WS3 handoff findings in comprehensive documentation.
Out of Scope Changes check	✅ Passed	The pull request contains only documentation additions scoped to the design note and includes no code changes, implementation, or upstream modifications, staying within declared non-goals.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch docs/ws5-vime-hook-points

---

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

[coderabbitai]

🧹 Nitpick comments (1)

docs/design/vime-architecture-and-hook-points.md (1)

237-244: ⚡ Quick win

Non-Goals section is too repetitive; rephrase to vary sentence openings.

Three successive bullet points all begin with "No", which is stylistically awkward and was flagged by static analysis. Rephrasing to vary the sentence structure will improve readability.

📝 Proposed rephrase to reduce repetition

 ## Non-Goals

-- No RL-Kernel operator injection.
-- No Vime, slime, vLLM, or Megatron source changes.
-- No production integration path.
-- No numerical claim that real vLLM and real Megatron are aligned.
-- No performance benchmark.
+- RL-Kernel operator injection is out of scope.
+- Source changes to Vime, slime, vLLM, or Megatron are not included.
+- Production integration pathways are deferred.
+- Numerical alignment claims between vLLM and Megatron are not made.
+- Performance benchmarking is excluded.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/design/vime-architecture-and-hook-points.md` around lines 237 - 244, The
Non-Goals section contains five consecutive bullet points that all begin with
"No", creating stylistic repetition that impacts readability. Rephrase these
bullet points to vary the sentence structure and openings while maintaining the
same meaning and intent. Consider using different grammatical constructions such
as passive voice, alternate phrasings, or restructuring how each point is
expressed. The goal is to break up the monotonous "No" pattern at the start of
each bullet while keeping the content clear and concise.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@docs/design/vime-architecture-and-hook-points.md`:
- Around line 237-244: The Non-Goals section contains five consecutive bullet
points that all begin with "No", creating stylistic repetition that impacts
readability. Rephrase these bullet points to vary the sentence structure and
openings while maintaining the same meaning and intent. Consider using different
grammatical constructions such as passive voice, alternate phrasings, or
restructuring how each point is expressed. The goal is to break up the
monotonous "No" pattern at the start of each bullet while keeping the content
clear and concise.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: c7ba5abf-478a-4533-95ce-8b45f67dacd3

📥 Commits

Reviewing files that changed from the base of the PR and between e3446bc and 2dcfaee.

📒 Files selected for processing (1)

• docs/design/vime-architecture-and-hook-points.md