The junior-v2 interview is showing it's age, I created it back when llama was all we had and at the time every single open source model failed the test.
The clustering we now see at the top of the leaderboard is a result of the massive improvements in open source coding models these past 6 months, anything above .95 is a binary pass and junior-v2 has no comparing ability up here.
The junior-v2 interview is showing it's age, I created it back when llama was all we had and at the time every single open source model failed the test.
The clustering we now see at the top of the leaderboard is a result of the massive improvements in open source coding models these past 6 months, anything above .95 is a binary pass and junior-v2 has no comparing ability up here.
A more difficult test suite is needed.