feat(toki): persistent SQLite leaderboard + API + live demo page (v0.10.0)#2
Merged
Merged
Conversation
…10.0)
T3 — give toki a leaderboard that survives across runs and tracks
robustness over time, surfaced as a live HTML page.
New module — toki.leaderboard (pure stdlib sqlite3):
- LeaderboardEntry(model_name, suite, pass_rate, robustness_score,
timestamp, notes, id) with [0,1] range + NaN + non-empty validation
at the boundary.
- Leaderboard.record / record_many / top_n / history / compare / all /
count. top_n("all") drops the suite filter for global ranking.
compare() picks latest-per-suite for both models and resolves a
winner over the overlapping suites.
- Schema auto-creates on first connection; KNOWN_SUITES =
(adversarial, paraphrase, noise) is the public contract for tabs.
Demo wiring (demo/server.py):
- POST /api/leaderboard, GET /api/leaderboard/{suite}, GET
/api/leaderboard/model/{name}. Lazy-init singleton; auto-seeds
from demo/seed_leaderboard.json on empty DB.
- demo/leaderboard.html — live page with suite filter tabs, 10s
auto-refresh, color-coded scores (green ≥0.85, yellow ≥0.70,
red <0.70), row-flash on new rows, offline indicator.
- demo/seed_leaderboard.json — 8 entries across 4 models
(phi-3, qwen-2.5, llama-3.2, gemma-2) and all 3 suites.
Breaking rename — Phase-7 leaderboard → ranking:
The Phase-7 module was a one-shot k-model Bonferroni-corrected
*ranking* operation, not a persistent leaderboard. Renaming it
frees the leaderboard namespace for what T3 actually needs.
- toki.leaderboard (Phase 7) → toki.ranking
- Leaderboard/Config/Entry/Result → Ranking/Config/Entry/Result
- CLI: python -m toki leaderboard → python -m toki rank
- Save artefact leaderboard.json → ranking.json
- Default --output-dir experiments/leaderboards →
experiments/rankings
- toki.__init__ exports updated; no compatibility shim (pre-1.0)
Tests:
- python/tests/test_leaderboard.py — 10 new tests covering schema
+ empty reads, record round-trip + auto-id, top_n sort/cap/filter,
top_n("all") global, chronological history, compare latest-per-
suite + winner, score range validation, load_seed bulk insert,
cross-instance persistence + KNOWN_SUITES contract.
- test_main.py CLI tests renamed (leaderboard → rank), assert
ranking.json artefact.
- test_ranking.py — Phase-7's 20 tests carried over verbatim.
Full test suite: 189 Python tests pass, 7 Rust tests pass, cargo
clippy clean. demo/leaderboard.db is gitignored.
Versions bumped: pyproject.toml + toki/__init__.py → 0.10.0.
Co-Authored-By: Claude Opus 4.7 (1M context) <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
2 participants
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
T3 — adds a persistent leaderboard that tracks model robustness scores over time, exposed via three new HTTP endpoints and a live HTML page.
toki/leaderboard.py— pure-stdlibsqlite3module.LeaderboardEntrydataclass +Leaderboardclass withrecord,top_n,history,compare. Schema auto-creates on first use; values validated at the unit-interval boundary.demo/server.py):POST /api/leaderboard,GET /api/leaderboard/{suite},GET /api/leaderboard/model/{name}. Suite ∈adversarial | paraphrase | noise | all.demo/leaderboard.html— live page at/leaderboard.html. Suite filter tabs, 10s auto-refresh, score colour-coding (green ≥0.85 / yellow ≥0.70 / red <0.70), row-flash on new entries, offline indicator.demo/seed_leaderboard.json— 8 entries across phi-3-mini-4k, qwen-2.5-1.5b, llama-3.2-3b, gemma-2-2b and all three suites; auto-loaded on first request.Breaking rename: Phase-7 leaderboard → ranking
The Phase-7 module was a one-shot k-model Bonferroni-corrected ranking operation, not a persistent leaderboard. To keep the namespace honest:
toki.leaderboard(Phase 7) →toki.rankingLeaderboard / Config / Entry / Result→Ranking / Config / Entry / Resultpython -m toki leaderboard→python -m toki rankleaderboard.json→ranking.json--output-dir:experiments/leaderboards→experiments/rankingsNo compatibility shim (pre-1.0). All Phase-7 tests carried over verbatim under the new module name and still pass.
Test plan
python -m pytest python/tests/— 189 passed (10 new intest_leaderboard.py, 20 carried over intest_ranking.py, CLI tests renamed)cargo test— 7 passedcargo clippy -- -D warnings— cleandemo/server.py(seed loaded, POST records, suite + model GETs return expected shape)/leaderboard.htmlin a browser — table renders, tabs filter, refresh ticks, error banner shows when server is down🤖 Generated with Claude Code