You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We've now had a closer a look. It appears that mainly base models, which have not been instruction-tuned, fail to follow the instructions to reason step by step, i.e. fail to generate reasoning traces at all. – Which makes sense.
ggbetz
changed the title
Why are many CoT reasoning traces empty
Why are many CoT reasoning traces empty?
Apr 17, 2024
In the DatasetViewer of
cot-leaderboard/cot-eval-traces-2.0
, it appears that many reasoning traces. Is this a bug?The text was updated successfully, but these errors were encountered: