-
Bug
-
Resolution: Done
-
Critical
-
None
-
None
-
False
-
-
False
-
-
-
Proposed
meyceoz discovered the precomputed dataset contains 18 samples which reference 7B. This leads to data output from sdg with training samples also containing 7B. This may lead to a model mis-indentifying itself as 7b instead of 8b.
- mentioned on
(6 mentioned on)