same-data-different-losses / _src /online_accuracy_grid.json
AlexWortega's picture
Interactive blog: Same Data, Different Losses, Same Circuits?
378a074 verified
Raw
History Blame Contribute Delete
373 Bytes
{"_about":"Online GRPO/DAPO greedy pass@1, gsm8k@1024 / aime26@2048, lr x seed grid.",
"grpo":{"5e-7":{"42":[93.3,13.3],"123":[93.1,16.7]},"5e-6":{"42":[93.7,20.0],"123":[93.0,16.7]},"5e-5":{"42":[92.6,13.3],"123":[93.0,10.0]}},
"dapo":{"5e-7":{"42":[92.5,20.0],"123":[93.0,13.3]},"5e-6":{"42":[93.3,16.7],"123":[92.8,13.3]},"5e-5":{"42":[91.1,13.3],"123":[92.2,13.3]}}}