JevMade Sign in
← Back to experiments

Benchmarks & research

jev-tmmluplus-eval

Puts the Traditional Chinese TMMLU+ exam to Jev, one four-way question per item, and scores it the way the official leaderboard does.

Source screenshot of jev-tmmluplus-eval
SOURCE SCREENSHOTFull screenshot ↗

What it does

The four options are the response format, so a stray answer is impossible and every question parses cleanly.

How you can use it

Choose which subjects from the 66-topic Traditional Chinese exam you want to test. A developer can run the test tool for you using an account key from TypeSafe. The tool gives the model each question and forces it to pick one of four answers directly.

Because the system only sends back a chosen letter and chance numbers, nobody has to clean up chatty writing or long explanations. Remember that this tool only grades the built-in exam questions, not your own custom quizzes or written essays.

Maker-reported (not independently measured by JevMade): 77.57 (ievals category macro) · 78.66 pooled · 19,646 questions · 66/66 subjects · 0 API errors · $0.42, 91 seconds at --concurrency 64

Primitives
choice
Platform
Python
Added
Project created

Keep this for later

Sign in to bookmark experiments, guides and videos, and keep notes only you can see.

Continue to sign in

We’ll bring you back to this listing.