pia mertens

piamertens

1 1 24

AI & ML interests

None yet

Recent Activity

liked a model 3 days ago

ConwayResearch/Underdog-Woof

new activity 7 days ago

mradermacher/KnowLine-4B-Gen5.5-GGUF:How to use as decision model in llama.cpp

upvoted a paper about 1 month ago

CORE: Contrastive Reflection Enables Rapid Improvements in Reasoning

View all activity

Organizations

None yet

liked a model 3 days ago

ConwayResearch/Underdog-Woof

Text-to-Speech • Updated 9 days ago • 10 • 270

New activity in mradermacher/KnowLine-4B-Gen5.5-GGUF 7 days ago

How to use as decision model in llama.cpp

#2 opened 7 days ago by

piamertens

upvoted a paper about 1 month ago

CORE: Contrastive Reflection Enables Rapid Improvements in Reasoning

Paper • 2605.28742 • Published May 27 • 6

liked a model about 1 month ago

chtisgit/Qwen3.6-35B-A3B-2Bit-GSQ-hybrid-GGUF

Image-Text-to-Text • 35B • Updated 24 days ago • 8.17k • 20

liked a model about 2 months ago

AlexWortega/openjev

Text Classification • Updated Oct 1 • 654

reacted to mayafree's post with ❤️ about 2 months ago

Post

3,550

JEV Ecosystems — every answer-verification vendor publishes a benchmark, and every one of them wins it. So we ran 13 of them on one test set: 2,018 items, identical labels, same grading code.

🎯 Leaderboard
mayafree/typed-decision-leaderboard

📄 Full write-up (method, mechanism, limits)
https://huggingface.co/blog/mayafree/jve-ecosystems

🧪 Try it — ZTC, JEV and Laya on the same input, side by side
mayafree/verifier-playground

Three results

1️⃣ Only three systems clear 0.70 — ZTC (397B) 0.7364 · JEV 0.7350 · ZTC (27B) 0.7282. First and second differ by 0.0014, so no rank is assigned.

2️⃣ A baseline that reads nothing but answer length and formatting scores 0.7036. Eight of the thirteen fall below it. A leaderboard without that line is flattering its entrants.

3️⃣ Bigger does not win. On scientific reasoning, 27B 0.7410 beats 397B 0.6287 — a model fourteen times larger scoring 0.11 lower.

And AUC is not the number you deploy on.

Same 20% retry budget, wired into an agent loop, against a 74.83% no-gate baseline:
ZTC +1.34 pp · JEV −0.07 pp · random −0.25 pp.

The mechanism is the interesting part. Re-answering is double-edged: 38% of wrong answers get fixed, and 30% of right answers get broken. So a gate is paid for by precision, not recall. Of the 403 items JEV routed for a retry, 216 were already correct.

0.0014 AUC apart; 1.4 points of end-to-end agent accuracy apart.

Scores, labels and grading code are published in full. Four public reproductions that would not run from their released artefacts are listed too, with the failure and a link, and no score.

Don't take the table's word for it — paste your own case into the playground and watch all three answer at once. Want a system added? Open a discussion on the Space.