feat(grpo): GRPO-suitability tagging per Autodata 2606.25996#85
Open
AdairBear wants to merge 4 commits into
Open
feat(grpo): GRPO-suitability tagging per Autodata 2606.25996#85AdairBear wants to merge 4 commits into
AdairBear wants to merge 4 commits into
Conversation
) QuantMind v0.2 ships ingestion + LLM extraction only; its persistence, embedding, semantic-query, and Data-MCP layers are unbuilt future PRs. This adds that missing Stage-2 layer as a self-contained package that reuses QuantMind's own venv and fetch+format layer: - store.py filesystem CorpusStore (JSON + .npy vectors, stable-hash dedup) - embed.py OpenAI embeddings + grounded answer synthesis + summarizer - ingest.py fetch_arxiv/url/local -> markdown -> summarize -> embed -> store (skips the brittle paper_flow Paper-tree: gpt-4o-mini emits non-UUID node ids that the Paper schema rejects) - query.py embed question -> cosine top-k -> grounded, cited answer - server.py FastMCP stdio server: qm_ingest_arxiv/url/pdf/text, qm_query, qm_list_corpus, qm_delete_item - cli.py seeding + shell use; seed_corpus.txt; _smoke_mcp.py handshake test Secrets load from ~/.hermes/.env; uses VOICE_TOOLS_OPENAI_KEY (real OpenAI) since Hermes OPENAI_API_KEY is an OpenRouter key with no embeddings endpoint. Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds `grpo_suitability: high|medium|low` to every corpus entry at ingest time, implementing the weak-vs-strong discrimination-gap framework from Kulikov et al. (FAIR at Meta, arXiv:2606.25996). V1 is a pure deterministic heuristic (no live model calls): - long + arxiv source + code present → high - short + news/unknown source + no code → low - everything else → medium Changes: - qm_mcp/grpo_suitability.py: GrpoSuitabilityScorer with score_entry(), length_band, domain_band, code_present helpers; V2 solver-gap hooks documented as TODOs - qm_mcp/ingest.py: score computed in _persist() and persisted to both items/<id>.json and ingestion_log.jsonl; backward-compatible (existing entries not touched) - qm_mcp/test_grpo_suitability.py: 22 pytest cases covering heuristic correctness, domain-band edge cases, backward compat, idempotency - docs/grpo_suitability.md: framework reference, V1 rule table, V2 plan Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Keeps the coverage floor enforced by CI (scripts/verify.sh) while allowing sub-package test suites (e.g. qm_mcp/) to run standalone without a false failure when quantmind code is not exercised. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
grpo_suitability: high|medium|lowto every corpus entry at ingest time, implementing the weak-vs-strong discrimination-gap framework from Kulikov et al. (FAIR at Meta, arXiv:2606.25996)null) on legacy recordsChanges
qm_mcp/grpo_suitability.pyGrpoSuitabilityScorerwithscore_entry(), length/domain/code helpers; V2 TODO hooksqm_mcp/ingest.py_persist(), persisted to item JSON + ingestion logqm_mcp/test_grpo_suitability.pydocs/grpo_suitability.mdV1 Heuristic Rule
Test Plan
pytest qm_mcp/test_grpo_suitability.py -v— 22/22 passed🤖 Generated with Claude Code