Dumb 1.2 Preview 0625(BASE)
#30
by 56m - opened
https://huggingface.co/56m/Dumb-1.2-Preview-0625
arc_easy(acc_norm): 0.3279
wikitext(byte / word ppl): 2.875 / 283.48
BLiMP(acc_norm): 0.6405
training token: est. 115~165M(dumb 1.2 exp 0616/0618/pre 0625)
param: 34.6M
Would you say this is a good model?
Also, please be sure to add this AI to the leaderboard.
Yeah I'll add it in a bit
Done!
Dumb-1.2-Preview-0625 is 18th
CompactAI changed discussion status to closed