Commit
•
0fae5dc
1
Parent(s):
303c73a
Upload 2 files
Browse files
results_2024-05-27T06-04-08.229332.json
ADDED
|
@@ -0,0 +1,1409 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"config_general": {
|
| 3 |
+
"lighteval_sha": "494ee12240e716e804ae9ea834f84a2c864c07ca",
|
| 4 |
+
"num_few_shot_default": 0,
|
| 5 |
+
"num_fewshot_seeds": 1,
|
| 6 |
+
"override_batch_size": 1,
|
| 7 |
+
"max_samples": null,
|
| 8 |
+
"job_id": "",
|
| 9 |
+
"start_time": 2370587.736800548,
|
| 10 |
+
"end_time": 2372824.830827315,
|
| 11 |
+
"total_evaluation_time_secondes": "2237.0940267667174",
|
| 12 |
+
"model_name": "Ramikan-BR/tinyllama-coder-py-v12",
|
| 13 |
+
"model_sha": "5835856d42314f549c92bb77eb9ca3e44edd1cda",
|
| 14 |
+
"model_dtype": "torch.float16",
|
| 15 |
+
"model_size": "2.05 GB"
|
| 16 |
+
},
|
| 17 |
+
"results": {
|
| 18 |
+
"harness|arc:challenge|25": {
|
| 19 |
+
"acc": 0.2858361774744027,
|
| 20 |
+
"acc_stderr": 0.01320319608853737,
|
| 21 |
+
"acc_norm": 0.3199658703071672,
|
| 22 |
+
"acc_norm_stderr": 0.013631345807016193
|
| 23 |
+
},
|
| 24 |
+
"harness|hellaswag|10": {
|
| 25 |
+
"acc": 0.41276638119896436,
|
| 26 |
+
"acc_stderr": 0.004913253031155681,
|
| 27 |
+
"acc_norm": 0.5361481776538538,
|
| 28 |
+
"acc_norm_stderr": 0.004976724124850562
|
| 29 |
+
},
|
| 30 |
+
"harness|hendrycksTest-abstract_algebra|5": {
|
| 31 |
+
"acc": 0.23,
|
| 32 |
+
"acc_stderr": 0.04229525846816505,
|
| 33 |
+
"acc_norm": 0.23,
|
| 34 |
+
"acc_norm_stderr": 0.04229525846816505
|
| 35 |
+
},
|
| 36 |
+
"harness|hendrycksTest-anatomy|5": {
|
| 37 |
+
"acc": 0.2814814814814815,
|
| 38 |
+
"acc_stderr": 0.038850042458002526,
|
| 39 |
+
"acc_norm": 0.2814814814814815,
|
| 40 |
+
"acc_norm_stderr": 0.038850042458002526
|
| 41 |
+
},
|
| 42 |
+
"harness|hendrycksTest-astronomy|5": {
|
| 43 |
+
"acc": 0.17763157894736842,
|
| 44 |
+
"acc_stderr": 0.031103182383123387,
|
| 45 |
+
"acc_norm": 0.17763157894736842,
|
| 46 |
+
"acc_norm_stderr": 0.031103182383123387
|
| 47 |
+
},
|
| 48 |
+
"harness|hendrycksTest-business_ethics|5": {
|
| 49 |
+
"acc": 0.26,
|
| 50 |
+
"acc_stderr": 0.044084400227680794,
|
| 51 |
+
"acc_norm": 0.26,
|
| 52 |
+
"acc_norm_stderr": 0.044084400227680794
|
| 53 |
+
},
|
| 54 |
+
"harness|hendrycksTest-clinical_knowledge|5": {
|
| 55 |
+
"acc": 0.24528301886792453,
|
| 56 |
+
"acc_stderr": 0.026480357179895685,
|
| 57 |
+
"acc_norm": 0.24528301886792453,
|
| 58 |
+
"acc_norm_stderr": 0.026480357179895685
|
| 59 |
+
},
|
| 60 |
+
"harness|hendrycksTest-college_biology|5": {
|
| 61 |
+
"acc": 0.2361111111111111,
|
| 62 |
+
"acc_stderr": 0.03551446610810826,
|
| 63 |
+
"acc_norm": 0.2361111111111111,
|
| 64 |
+
"acc_norm_stderr": 0.03551446610810826
|
| 65 |
+
},
|
| 66 |
+
"harness|hendrycksTest-college_chemistry|5": {
|
| 67 |
+
"acc": 0.39,
|
| 68 |
+
"acc_stderr": 0.04902071300001974,
|
| 69 |
+
"acc_norm": 0.39,
|
| 70 |
+
"acc_norm_stderr": 0.04902071300001974
|
| 71 |
+
},
|
| 72 |
+
"harness|hendrycksTest-college_computer_science|5": {
|
| 73 |
+
"acc": 0.33,
|
| 74 |
+
"acc_stderr": 0.04725815626252604,
|
| 75 |
+
"acc_norm": 0.33,
|
| 76 |
+
"acc_norm_stderr": 0.04725815626252604
|
| 77 |
+
},
|
| 78 |
+
"harness|hendrycksTest-college_mathematics|5": {
|
| 79 |
+
"acc": 0.33,
|
| 80 |
+
"acc_stderr": 0.047258156262526045,
|
| 81 |
+
"acc_norm": 0.33,
|
| 82 |
+
"acc_norm_stderr": 0.047258156262526045
|
| 83 |
+
},
|
| 84 |
+
"harness|hendrycksTest-college_medicine|5": {
|
| 85 |
+
"acc": 0.2138728323699422,
|
| 86 |
+
"acc_stderr": 0.03126511206173043,
|
| 87 |
+
"acc_norm": 0.2138728323699422,
|
| 88 |
+
"acc_norm_stderr": 0.03126511206173043
|
| 89 |
+
},
|
| 90 |
+
"harness|hendrycksTest-college_physics|5": {
|
| 91 |
+
"acc": 0.21568627450980393,
|
| 92 |
+
"acc_stderr": 0.04092563958237655,
|
| 93 |
+
"acc_norm": 0.21568627450980393,
|
| 94 |
+
"acc_norm_stderr": 0.04092563958237655
|
| 95 |
+
},
|
| 96 |
+
"harness|hendrycksTest-computer_security|5": {
|
| 97 |
+
"acc": 0.31,
|
| 98 |
+
"acc_stderr": 0.04648231987117316,
|
| 99 |
+
"acc_norm": 0.31,
|
| 100 |
+
"acc_norm_stderr": 0.04648231987117316
|
| 101 |
+
},
|
| 102 |
+
"harness|hendrycksTest-conceptual_physics|5": {
|
| 103 |
+
"acc": 0.251063829787234,
|
| 104 |
+
"acc_stderr": 0.02834696377716245,
|
| 105 |
+
"acc_norm": 0.251063829787234,
|
| 106 |
+
"acc_norm_stderr": 0.02834696377716245
|
| 107 |
+
},
|
| 108 |
+
"harness|hendrycksTest-econometrics|5": {
|
| 109 |
+
"acc": 0.2894736842105263,
|
| 110 |
+
"acc_stderr": 0.04266339443159394,
|
| 111 |
+
"acc_norm": 0.2894736842105263,
|
| 112 |
+
"acc_norm_stderr": 0.04266339443159394
|
| 113 |
+
},
|
| 114 |
+
"harness|hendrycksTest-electrical_engineering|5": {
|
| 115 |
+
"acc": 0.21379310344827587,
|
| 116 |
+
"acc_stderr": 0.03416520447747548,
|
| 117 |
+
"acc_norm": 0.21379310344827587,
|
| 118 |
+
"acc_norm_stderr": 0.03416520447747548
|
| 119 |
+
},
|
| 120 |
+
"harness|hendrycksTest-elementary_mathematics|5": {
|
| 121 |
+
"acc": 0.2222222222222222,
|
| 122 |
+
"acc_stderr": 0.021411684393694203,
|
| 123 |
+
"acc_norm": 0.2222222222222222,
|
| 124 |
+
"acc_norm_stderr": 0.021411684393694203
|
| 125 |
+
},
|
| 126 |
+
"harness|hendrycksTest-formal_logic|5": {
|
| 127 |
+
"acc": 0.21428571428571427,
|
| 128 |
+
"acc_stderr": 0.03670066451047181,
|
| 129 |
+
"acc_norm": 0.21428571428571427,
|
| 130 |
+
"acc_norm_stderr": 0.03670066451047181
|
| 131 |
+
},
|
| 132 |
+
"harness|hendrycksTest-global_facts|5": {
|
| 133 |
+
"acc": 0.3,
|
| 134 |
+
"acc_stderr": 0.046056618647183814,
|
| 135 |
+
"acc_norm": 0.3,
|
| 136 |
+
"acc_norm_stderr": 0.046056618647183814
|
| 137 |
+
},
|
| 138 |
+
"harness|hendrycksTest-high_school_biology|5": {
|
| 139 |
+
"acc": 0.2064516129032258,
|
| 140 |
+
"acc_stderr": 0.023025899617188726,
|
| 141 |
+
"acc_norm": 0.2064516129032258,
|
| 142 |
+
"acc_norm_stderr": 0.023025899617188726
|
| 143 |
+
},
|
| 144 |
+
"harness|hendrycksTest-high_school_chemistry|5": {
|
| 145 |
+
"acc": 0.1625615763546798,
|
| 146 |
+
"acc_stderr": 0.025960300064605597,
|
| 147 |
+
"acc_norm": 0.1625615763546798,
|
| 148 |
+
"acc_norm_stderr": 0.025960300064605597
|
| 149 |
+
},
|
| 150 |
+
"harness|hendrycksTest-high_school_computer_science|5": {
|
| 151 |
+
"acc": 0.3,
|
| 152 |
+
"acc_stderr": 0.046056618647183814,
|
| 153 |
+
"acc_norm": 0.3,
|
| 154 |
+
"acc_norm_stderr": 0.046056618647183814
|
| 155 |
+
},
|
| 156 |
+
"harness|hendrycksTest-high_school_european_history|5": {
|
| 157 |
+
"acc": 0.21212121212121213,
|
| 158 |
+
"acc_stderr": 0.03192271569548299,
|
| 159 |
+
"acc_norm": 0.21212121212121213,
|
| 160 |
+
"acc_norm_stderr": 0.03192271569548299
|
| 161 |
+
},
|
| 162 |
+
"harness|hendrycksTest-high_school_geography|5": {
|
| 163 |
+
"acc": 0.20707070707070707,
|
| 164 |
+
"acc_stderr": 0.028869778460267063,
|
| 165 |
+
"acc_norm": 0.20707070707070707,
|
| 166 |
+
"acc_norm_stderr": 0.028869778460267063
|
| 167 |
+
},
|
| 168 |
+
"harness|hendrycksTest-high_school_government_and_politics|5": {
|
| 169 |
+
"acc": 0.26424870466321243,
|
| 170 |
+
"acc_stderr": 0.031821550509166484,
|
| 171 |
+
"acc_norm": 0.26424870466321243,
|
| 172 |
+
"acc_norm_stderr": 0.031821550509166484
|
| 173 |
+
},
|
| 174 |
+
"harness|hendrycksTest-high_school_macroeconomics|5": {
|
| 175 |
+
"acc": 0.32051282051282054,
|
| 176 |
+
"acc_stderr": 0.023661296393964273,
|
| 177 |
+
"acc_norm": 0.32051282051282054,
|
| 178 |
+
"acc_norm_stderr": 0.023661296393964273
|
| 179 |
+
},
|
| 180 |
+
"harness|hendrycksTest-high_school_mathematics|5": {
|
| 181 |
+
"acc": 0.2814814814814815,
|
| 182 |
+
"acc_stderr": 0.027420019350945277,
|
| 183 |
+
"acc_norm": 0.2814814814814815,
|
| 184 |
+
"acc_norm_stderr": 0.027420019350945277
|
| 185 |
+
},
|
| 186 |
+
"harness|hendrycksTest-high_school_microeconomics|5": {
|
| 187 |
+
"acc": 0.22268907563025211,
|
| 188 |
+
"acc_stderr": 0.027025433498882367,
|
| 189 |
+
"acc_norm": 0.22268907563025211,
|
| 190 |
+
"acc_norm_stderr": 0.027025433498882367
|
| 191 |
+
},
|
| 192 |
+
"harness|hendrycksTest-high_school_physics|5": {
|
| 193 |
+
"acc": 0.2847682119205298,
|
| 194 |
+
"acc_stderr": 0.03684881521389023,
|
| 195 |
+
"acc_norm": 0.2847682119205298,
|
| 196 |
+
"acc_norm_stderr": 0.03684881521389023
|
| 197 |
+
},
|
| 198 |
+
"harness|hendrycksTest-high_school_psychology|5": {
|
| 199 |
+
"acc": 0.24954128440366974,
|
| 200 |
+
"acc_stderr": 0.01855389762950161,
|
| 201 |
+
"acc_norm": 0.24954128440366974,
|
| 202 |
+
"acc_norm_stderr": 0.01855389762950161
|
| 203 |
+
},
|
| 204 |
+
"harness|hendrycksTest-high_school_statistics|5": {
|
| 205 |
+
"acc": 0.4722222222222222,
|
| 206 |
+
"acc_stderr": 0.0340470532865388,
|
| 207 |
+
"acc_norm": 0.4722222222222222,
|
| 208 |
+
"acc_norm_stderr": 0.0340470532865388
|
| 209 |
+
},
|
| 210 |
+
"harness|hendrycksTest-high_school_us_history|5": {
|
| 211 |
+
"acc": 0.25980392156862747,
|
| 212 |
+
"acc_stderr": 0.030778554678693264,
|
| 213 |
+
"acc_norm": 0.25980392156862747,
|
| 214 |
+
"acc_norm_stderr": 0.030778554678693264
|
| 215 |
+
},
|
| 216 |
+
"harness|hendrycksTest-high_school_world_history|5": {
|
| 217 |
+
"acc": 0.2616033755274262,
|
| 218 |
+
"acc_stderr": 0.028609516716994934,
|
| 219 |
+
"acc_norm": 0.2616033755274262,
|
| 220 |
+
"acc_norm_stderr": 0.028609516716994934
|
| 221 |
+
},
|
| 222 |
+
"harness|hendrycksTest-human_aging|5": {
|
| 223 |
+
"acc": 0.3004484304932735,
|
| 224 |
+
"acc_stderr": 0.030769352008229143,
|
| 225 |
+
"acc_norm": 0.3004484304932735,
|
| 226 |
+
"acc_norm_stderr": 0.030769352008229143
|
| 227 |
+
},
|
| 228 |
+
"harness|hendrycksTest-human_sexuality|5": {
|
| 229 |
+
"acc": 0.26717557251908397,
|
| 230 |
+
"acc_stderr": 0.038808483010823944,
|
| 231 |
+
"acc_norm": 0.26717557251908397,
|
| 232 |
+
"acc_norm_stderr": 0.038808483010823944
|
| 233 |
+
},
|
| 234 |
+
"harness|hendrycksTest-international_law|5": {
|
| 235 |
+
"acc": 0.24793388429752067,
|
| 236 |
+
"acc_stderr": 0.03941897526516302,
|
| 237 |
+
"acc_norm": 0.24793388429752067,
|
| 238 |
+
"acc_norm_stderr": 0.03941897526516302
|
| 239 |
+
},
|
| 240 |
+
"harness|hendrycksTest-jurisprudence|5": {
|
| 241 |
+
"acc": 0.25925925925925924,
|
| 242 |
+
"acc_stderr": 0.042365112580946336,
|
| 243 |
+
"acc_norm": 0.25925925925925924,
|
| 244 |
+
"acc_norm_stderr": 0.042365112580946336
|
| 245 |
+
},
|
| 246 |
+
"harness|hendrycksTest-logical_fallacies|5": {
|
| 247 |
+
"acc": 0.2883435582822086,
|
| 248 |
+
"acc_stderr": 0.035590395316173425,
|
| 249 |
+
"acc_norm": 0.2883435582822086,
|
| 250 |
+
"acc_norm_stderr": 0.035590395316173425
|
| 251 |
+
},
|
| 252 |
+
"harness|hendrycksTest-machine_learning|5": {
|
| 253 |
+
"acc": 0.25892857142857145,
|
| 254 |
+
"acc_stderr": 0.041577515398656284,
|
| 255 |
+
"acc_norm": 0.25892857142857145,
|
| 256 |
+
"acc_norm_stderr": 0.041577515398656284
|
| 257 |
+
},
|
| 258 |
+
"harness|hendrycksTest-management|5": {
|
| 259 |
+
"acc": 0.1650485436893204,
|
| 260 |
+
"acc_stderr": 0.036756688322331886,
|
| 261 |
+
"acc_norm": 0.1650485436893204,
|
| 262 |
+
"acc_norm_stderr": 0.036756688322331886
|
| 263 |
+
},
|
| 264 |
+
"harness|hendrycksTest-marketing|5": {
|
| 265 |
+
"acc": 0.3247863247863248,
|
| 266 |
+
"acc_stderr": 0.030679022765498835,
|
| 267 |
+
"acc_norm": 0.3247863247863248,
|
| 268 |
+
"acc_norm_stderr": 0.030679022765498835
|
| 269 |
+
},
|
| 270 |
+
"harness|hendrycksTest-medical_genetics|5": {
|
| 271 |
+
"acc": 0.29,
|
| 272 |
+
"acc_stderr": 0.045604802157206845,
|
| 273 |
+
"acc_norm": 0.29,
|
| 274 |
+
"acc_norm_stderr": 0.045604802157206845
|
| 275 |
+
},
|
| 276 |
+
"harness|hendrycksTest-miscellaneous|5": {
|
| 277 |
+
"acc": 0.280970625798212,
|
| 278 |
+
"acc_stderr": 0.01607312785122126,
|
| 279 |
+
"acc_norm": 0.280970625798212,
|
| 280 |
+
"acc_norm_stderr": 0.01607312785122126
|
| 281 |
+
},
|
| 282 |
+
"harness|hendrycksTest-moral_disputes|5": {
|
| 283 |
+
"acc": 0.24855491329479767,
|
| 284 |
+
"acc_stderr": 0.023267528432100174,
|
| 285 |
+
"acc_norm": 0.24855491329479767,
|
| 286 |
+
"acc_norm_stderr": 0.023267528432100174
|
| 287 |
+
},
|
| 288 |
+
"harness|hendrycksTest-moral_scenarios|5": {
|
| 289 |
+
"acc": 0.25027932960893856,
|
| 290 |
+
"acc_stderr": 0.01448750085285041,
|
| 291 |
+
"acc_norm": 0.25027932960893856,
|
| 292 |
+
"acc_norm_stderr": 0.01448750085285041
|
| 293 |
+
},
|
| 294 |
+
"harness|hendrycksTest-nutrition|5": {
|
| 295 |
+
"acc": 0.21895424836601307,
|
| 296 |
+
"acc_stderr": 0.02367908986180772,
|
| 297 |
+
"acc_norm": 0.21895424836601307,
|
| 298 |
+
"acc_norm_stderr": 0.02367908986180772
|
| 299 |
+
},
|
| 300 |
+
"harness|hendrycksTest-philosophy|5": {
|
| 301 |
+
"acc": 0.2733118971061093,
|
| 302 |
+
"acc_stderr": 0.025311765975426115,
|
| 303 |
+
"acc_norm": 0.2733118971061093,
|
| 304 |
+
"acc_norm_stderr": 0.025311765975426115
|
| 305 |
+
},
|
| 306 |
+
"harness|hendrycksTest-prehistory|5": {
|
| 307 |
+
"acc": 0.19135802469135801,
|
| 308 |
+
"acc_stderr": 0.021887704613396158,
|
| 309 |
+
"acc_norm": 0.19135802469135801,
|
| 310 |
+
"acc_norm_stderr": 0.021887704613396158
|
| 311 |
+
},
|
| 312 |
+
"harness|hendrycksTest-professional_accounting|5": {
|
| 313 |
+
"acc": 0.24113475177304963,
|
| 314 |
+
"acc_stderr": 0.02551873104953776,
|
| 315 |
+
"acc_norm": 0.24113475177304963,
|
| 316 |
+
"acc_norm_stderr": 0.02551873104953776
|
| 317 |
+
},
|
| 318 |
+
"harness|hendrycksTest-professional_law|5": {
|
| 319 |
+
"acc": 0.24445893089960888,
|
| 320 |
+
"acc_stderr": 0.010976425013113897,
|
| 321 |
+
"acc_norm": 0.24445893089960888,
|
| 322 |
+
"acc_norm_stderr": 0.010976425013113897
|
| 323 |
+
},
|
| 324 |
+
"harness|hendrycksTest-professional_medicine|5": {
|
| 325 |
+
"acc": 0.4485294117647059,
|
| 326 |
+
"acc_stderr": 0.030211479609121593,
|
| 327 |
+
"acc_norm": 0.4485294117647059,
|
| 328 |
+
"acc_norm_stderr": 0.030211479609121593
|
| 329 |
+
},
|
| 330 |
+
"harness|hendrycksTest-professional_psychology|5": {
|
| 331 |
+
"acc": 0.21568627450980393,
|
| 332 |
+
"acc_stderr": 0.01663931935031326,
|
| 333 |
+
"acc_norm": 0.21568627450980393,
|
| 334 |
+
"acc_norm_stderr": 0.01663931935031326
|
| 335 |
+
},
|
| 336 |
+
"harness|hendrycksTest-public_relations|5": {
|
| 337 |
+
"acc": 0.23636363636363636,
|
| 338 |
+
"acc_stderr": 0.04069306319721376,
|
| 339 |
+
"acc_norm": 0.23636363636363636,
|
| 340 |
+
"acc_norm_stderr": 0.04069306319721376
|
| 341 |
+
},
|
| 342 |
+
"harness|hendrycksTest-security_studies|5": {
|
| 343 |
+
"acc": 0.2571428571428571,
|
| 344 |
+
"acc_stderr": 0.02797982353874455,
|
| 345 |
+
"acc_norm": 0.2571428571428571,
|
| 346 |
+
"acc_norm_stderr": 0.02797982353874455
|
| 347 |
+
},
|
| 348 |
+
"harness|hendrycksTest-sociology|5": {
|
| 349 |
+
"acc": 0.2537313432835821,
|
| 350 |
+
"acc_stderr": 0.030769444967296018,
|
| 351 |
+
"acc_norm": 0.2537313432835821,
|
| 352 |
+
"acc_norm_stderr": 0.030769444967296018
|
| 353 |
+
},
|
| 354 |
+
"harness|hendrycksTest-us_foreign_policy|5": {
|
| 355 |
+
"acc": 0.23,
|
| 356 |
+
"acc_stderr": 0.04229525846816506,
|
| 357 |
+
"acc_norm": 0.23,
|
| 358 |
+
"acc_norm_stderr": 0.04229525846816506
|
| 359 |
+
},
|
| 360 |
+
"harness|hendrycksTest-virology|5": {
|
| 361 |
+
"acc": 0.26506024096385544,
|
| 362 |
+
"acc_stderr": 0.03436024037944967,
|
| 363 |
+
"acc_norm": 0.26506024096385544,
|
| 364 |
+
"acc_norm_stderr": 0.03436024037944967
|
| 365 |
+
},
|
| 366 |
+
"harness|hendrycksTest-world_religions|5": {
|
| 367 |
+
"acc": 0.3216374269005848,
|
| 368 |
+
"acc_stderr": 0.03582529442573122,
|
| 369 |
+
"acc_norm": 0.3216374269005848,
|
| 370 |
+
"acc_norm_stderr": 0.03582529442573122
|
| 371 |
+
},
|
| 372 |
+
"harness|truthfulqa:mc|0": {
|
| 373 |
+
"mc1": 0.25458996328029376,
|
| 374 |
+
"mc1_stderr": 0.015250117079156496,
|
| 375 |
+
"mc2": 0.4091007666951377,
|
| 376 |
+
"mc2_stderr": 0.014365367143474025
|
| 377 |
+
},
|
| 378 |
+
"harness|winogrande|5": {
|
| 379 |
+
"acc": 0.5706393054459353,
|
| 380 |
+
"acc_stderr": 0.013911537499969163
|
| 381 |
+
},
|
| 382 |
+
"harness|gsm8k|5": {
|
| 383 |
+
"acc": 0.014404852160727824,
|
| 384 |
+
"acc_stderr": 0.0032820559171369444
|
| 385 |
+
},
|
| 386 |
+
"all": {
|
| 387 |
+
"acc": 0.26686435788728485,
|
| 388 |
+
"acc_stderr": 0.031169507718254618,
|
| 389 |
+
"acc_norm": 0.2686642908950062,
|
| 390 |
+
"acc_norm_stderr": 0.03194301691878982,
|
| 391 |
+
"mc1": 0.25458996328029376,
|
| 392 |
+
"mc1_stderr": 0.015250117079156496,
|
| 393 |
+
"mc2": 0.4091007666951377,
|
| 394 |
+
"mc2_stderr": 0.014365367143474025
|
| 395 |
+
}
|
| 396 |
+
},
|
| 397 |
+
"versions": {
|
| 398 |
+
"all": 0,
|
| 399 |
+
"harness|arc:challenge|25": 0,
|
| 400 |
+
"harness|gsm8k|5": 0,
|
| 401 |
+
"harness|hellaswag|10": 0,
|
| 402 |
+
"harness|hendrycksTest-abstract_algebra|5": 1,
|
| 403 |
+
"harness|hendrycksTest-anatomy|5": 1,
|
| 404 |
+
"harness|hendrycksTest-astronomy|5": 1,
|
| 405 |
+
"harness|hendrycksTest-business_ethics|5": 1,
|
| 406 |
+
"harness|hendrycksTest-clinical_knowledge|5": 1,
|
| 407 |
+
"harness|hendrycksTest-college_biology|5": 1,
|
| 408 |
+
"harness|hendrycksTest-college_chemistry|5": 1,
|
| 409 |
+
"harness|hendrycksTest-college_computer_science|5": 1,
|
| 410 |
+
"harness|hendrycksTest-college_mathematics|5": 1,
|
| 411 |
+
"harness|hendrycksTest-college_medicine|5": 1,
|
| 412 |
+
"harness|hendrycksTest-college_physics|5": 1,
|
| 413 |
+
"harness|hendrycksTest-computer_security|5": 1,
|
| 414 |
+
"harness|hendrycksTest-conceptual_physics|5": 1,
|
| 415 |
+
"harness|hendrycksTest-econometrics|5": 1,
|
| 416 |
+
"harness|hendrycksTest-electrical_engineering|5": 1,
|
| 417 |
+
"harness|hendrycksTest-elementary_mathematics|5": 1,
|
| 418 |
+
"harness|hendrycksTest-formal_logic|5": 1,
|
| 419 |
+
"harness|hendrycksTest-global_facts|5": 1,
|
| 420 |
+
"harness|hendrycksTest-high_school_biology|5": 1,
|
| 421 |
+
"harness|hendrycksTest-high_school_chemistry|5": 1,
|
| 422 |
+
"harness|hendrycksTest-high_school_computer_science|5": 1,
|
| 423 |
+
"harness|hendrycksTest-high_school_european_history|5": 1,
|
| 424 |
+
"harness|hendrycksTest-high_school_geography|5": 1,
|
| 425 |
+
"harness|hendrycksTest-high_school_government_and_politics|5": 1,
|
| 426 |
+
"harness|hendrycksTest-high_school_macroeconomics|5": 1,
|
| 427 |
+
"harness|hendrycksTest-high_school_mathematics|5": 1,
|
| 428 |
+
"harness|hendrycksTest-high_school_microeconomics|5": 1,
|
| 429 |
+
"harness|hendrycksTest-high_school_physics|5": 1,
|
| 430 |
+
"harness|hendrycksTest-high_school_psychology|5": 1,
|
| 431 |
+
"harness|hendrycksTest-high_school_statistics|5": 1,
|
| 432 |
+
"harness|hendrycksTest-high_school_us_history|5": 1,
|
| 433 |
+
"harness|hendrycksTest-high_school_world_history|5": 1,
|
| 434 |
+
"harness|hendrycksTest-human_aging|5": 1,
|
| 435 |
+
"harness|hendrycksTest-human_sexuality|5": 1,
|
| 436 |
+
"harness|hendrycksTest-international_law|5": 1,
|
| 437 |
+
"harness|hendrycksTest-jurisprudence|5": 1,
|
| 438 |
+
"harness|hendrycksTest-logical_fallacies|5": 1,
|
| 439 |
+
"harness|hendrycksTest-machine_learning|5": 1,
|
| 440 |
+
"harness|hendrycksTest-management|5": 1,
|
| 441 |
+
"harness|hendrycksTest-marketing|5": 1,
|
| 442 |
+
"harness|hendrycksTest-medical_genetics|5": 1,
|
| 443 |
+
"harness|hendrycksTest-miscellaneous|5": 1,
|
| 444 |
+
"harness|hendrycksTest-moral_disputes|5": 1,
|
| 445 |
+
"harness|hendrycksTest-moral_scenarios|5": 1,
|
| 446 |
+
"harness|hendrycksTest-nutrition|5": 1,
|
| 447 |
+
"harness|hendrycksTest-philosophy|5": 1,
|
| 448 |
+
"harness|hendrycksTest-prehistory|5": 1,
|
| 449 |
+
"harness|hendrycksTest-professional_accounting|5": 1,
|
| 450 |
+
"harness|hendrycksTest-professional_law|5": 1,
|
| 451 |
+
"harness|hendrycksTest-professional_medicine|5": 1,
|
| 452 |
+
"harness|hendrycksTest-professional_psychology|5": 1,
|
| 453 |
+
"harness|hendrycksTest-public_relations|5": 1,
|
| 454 |
+
"harness|hendrycksTest-security_studies|5": 1,
|
| 455 |
+
"harness|hendrycksTest-sociology|5": 1,
|
| 456 |
+
"harness|hendrycksTest-us_foreign_policy|5": 1,
|
| 457 |
+
"harness|hendrycksTest-virology|5": 1,
|
| 458 |
+
"harness|hendrycksTest-world_religions|5": 1,
|
| 459 |
+
"harness|truthfulqa:mc|0": 1,
|
| 460 |
+
"harness|winogrande|5": 0
|
| 461 |
+
},
|
| 462 |
+
"config_tasks": {
|
| 463 |
+
"harness|arc:challenge": "LM Harness task",
|
| 464 |
+
"harness|gsm8k": "LM Harness task",
|
| 465 |
+
"harness|hellaswag": "LM Harness task",
|
| 466 |
+
"harness|hendrycksTest-abstract_algebra": "LM Harness task",
|
| 467 |
+
"harness|hendrycksTest-anatomy": "LM Harness task",
|
| 468 |
+
"harness|hendrycksTest-astronomy": "LM Harness task",
|
| 469 |
+
"harness|hendrycksTest-business_ethics": "LM Harness task",
|
| 470 |
+
"harness|hendrycksTest-clinical_knowledge": "LM Harness task",
|
| 471 |
+
"harness|hendrycksTest-college_biology": "LM Harness task",
|
| 472 |
+
"harness|hendrycksTest-college_chemistry": "LM Harness task",
|
| 473 |
+
"harness|hendrycksTest-college_computer_science": "LM Harness task",
|
| 474 |
+
"harness|hendrycksTest-college_mathematics": "LM Harness task",
|
| 475 |
+
"harness|hendrycksTest-college_medicine": "LM Harness task",
|
| 476 |
+
"harness|hendrycksTest-college_physics": "LM Harness task",
|
| 477 |
+
"harness|hendrycksTest-computer_security": "LM Harness task",
|
| 478 |
+
"harness|hendrycksTest-conceptual_physics": "LM Harness task",
|
| 479 |
+
"harness|hendrycksTest-econometrics": "LM Harness task",
|
| 480 |
+
"harness|hendrycksTest-electrical_engineering": "LM Harness task",
|
| 481 |
+
"harness|hendrycksTest-elementary_mathematics": "LM Harness task",
|
| 482 |
+
"harness|hendrycksTest-formal_logic": "LM Harness task",
|
| 483 |
+
"harness|hendrycksTest-global_facts": "LM Harness task",
|
| 484 |
+
"harness|hendrycksTest-high_school_biology": "LM Harness task",
|
| 485 |
+
"harness|hendrycksTest-high_school_chemistry": "LM Harness task",
|
| 486 |
+
"harness|hendrycksTest-high_school_computer_science": "LM Harness task",
|
| 487 |
+
"harness|hendrycksTest-high_school_european_history": "LM Harness task",
|
| 488 |
+
"harness|hendrycksTest-high_school_geography": "LM Harness task",
|
| 489 |
+
"harness|hendrycksTest-high_school_government_and_politics": "LM Harness task",
|
| 490 |
+
"harness|hendrycksTest-high_school_macroeconomics": "LM Harness task",
|
| 491 |
+
"harness|hendrycksTest-high_school_mathematics": "LM Harness task",
|
| 492 |
+
"harness|hendrycksTest-high_school_microeconomics": "LM Harness task",
|
| 493 |
+
"harness|hendrycksTest-high_school_physics": "LM Harness task",
|
| 494 |
+
"harness|hendrycksTest-high_school_psychology": "LM Harness task",
|
| 495 |
+
"harness|hendrycksTest-high_school_statistics": "LM Harness task",
|
| 496 |
+
"harness|hendrycksTest-high_school_us_history": "LM Harness task",
|
| 497 |
+
"harness|hendrycksTest-high_school_world_history": "LM Harness task",
|
| 498 |
+
"harness|hendrycksTest-human_aging": "LM Harness task",
|
| 499 |
+
"harness|hendrycksTest-human_sexuality": "LM Harness task",
|
| 500 |
+
"harness|hendrycksTest-international_law": "LM Harness task",
|
| 501 |
+
"harness|hendrycksTest-jurisprudence": "LM Harness task",
|
| 502 |
+
"harness|hendrycksTest-logical_fallacies": "LM Harness task",
|
| 503 |
+
"harness|hendrycksTest-machine_learning": "LM Harness task",
|
| 504 |
+
"harness|hendrycksTest-management": "LM Harness task",
|
| 505 |
+
"harness|hendrycksTest-marketing": "LM Harness task",
|
| 506 |
+
"harness|hendrycksTest-medical_genetics": "LM Harness task",
|
| 507 |
+
"harness|hendrycksTest-miscellaneous": "LM Harness task",
|
| 508 |
+
"harness|hendrycksTest-moral_disputes": "LM Harness task",
|
| 509 |
+
"harness|hendrycksTest-moral_scenarios": "LM Harness task",
|
| 510 |
+
"harness|hendrycksTest-nutrition": "LM Harness task",
|
| 511 |
+
"harness|hendrycksTest-philosophy": "LM Harness task",
|
| 512 |
+
"harness|hendrycksTest-prehistory": "LM Harness task",
|
| 513 |
+
"harness|hendrycksTest-professional_accounting": "LM Harness task",
|
| 514 |
+
"harness|hendrycksTest-professional_law": "LM Harness task",
|
| 515 |
+
"harness|hendrycksTest-professional_medicine": "LM Harness task",
|
| 516 |
+
"harness|hendrycksTest-professional_psychology": "LM Harness task",
|
| 517 |
+
"harness|hendrycksTest-public_relations": "LM Harness task",
|
| 518 |
+
"harness|hendrycksTest-security_studies": "LM Harness task",
|
| 519 |
+
"harness|hendrycksTest-sociology": "LM Harness task",
|
| 520 |
+
"harness|hendrycksTest-us_foreign_policy": "LM Harness task",
|
| 521 |
+
"harness|hendrycksTest-virology": "LM Harness task",
|
| 522 |
+
"harness|hendrycksTest-world_religions": "LM Harness task",
|
| 523 |
+
"harness|truthfulqa:mc": "LM Harness task",
|
| 524 |
+
"harness|winogrande": "LM Harness task"
|
| 525 |
+
},
|
| 526 |
+
"summary_tasks": {
|
| 527 |
+
"harness|arc:challenge|25": {
|
| 528 |
+
"hashes": {
|
| 529 |
+
"hash_examples": "17b0cae357c0259e",
|
| 530 |
+
"hash_full_prompts": "045cbb916e5145c6",
|
| 531 |
+
"hash_input_tokens": "ca48d52265c0051f",
|
| 532 |
+
"hash_cont_tokens": "e8abf848493b50f7"
|
| 533 |
+
},
|
| 534 |
+
"truncated": 0,
|
| 535 |
+
"non_truncated": 1172,
|
| 536 |
+
"padded": 4687,
|
| 537 |
+
"non_padded": 0,
|
| 538 |
+
"effective_few_shots": 25.0,
|
| 539 |
+
"num_truncated_few_shots": 0
|
| 540 |
+
},
|
| 541 |
+
"harness|hellaswag|10": {
|
| 542 |
+
"hashes": {
|
| 543 |
+
"hash_examples": "e1768ecb99d7ecf0",
|
| 544 |
+
"hash_full_prompts": "0b4c16983130f84f",
|
| 545 |
+
"hash_input_tokens": "4975ded0ed31f702",
|
| 546 |
+
"hash_cont_tokens": "9fe0a5c42e1532db"
|
| 547 |
+
},
|
| 548 |
+
"truncated": 0,
|
| 549 |
+
"non_truncated": 10042,
|
| 550 |
+
"padded": 40019,
|
| 551 |
+
"non_padded": 149,
|
| 552 |
+
"effective_few_shots": 10.0,
|
| 553 |
+
"num_truncated_few_shots": 0
|
| 554 |
+
},
|
| 555 |
+
"harness|hendrycksTest-abstract_algebra|5": {
|
| 556 |
+
"hashes": {
|
| 557 |
+
"hash_examples": "280f9f325b40559a",
|
| 558 |
+
"hash_full_prompts": "2f776a367d23aea2",
|
| 559 |
+
"hash_input_tokens": "8ff523ec326d5d55",
|
| 560 |
+
"hash_cont_tokens": "50421e30bef398f9"
|
| 561 |
+
},
|
| 562 |
+
"truncated": 0,
|
| 563 |
+
"non_truncated": 100,
|
| 564 |
+
"padded": 400,
|
| 565 |
+
"non_padded": 0,
|
| 566 |
+
"effective_few_shots": 5.0,
|
| 567 |
+
"num_truncated_few_shots": 0
|
| 568 |
+
},
|
| 569 |
+
"harness|hendrycksTest-anatomy|5": {
|
| 570 |
+
"hashes": {
|
| 571 |
+
"hash_examples": "2f83a4f1cab4ba18",
|
| 572 |
+
"hash_full_prompts": "516f74bef25df620",
|
| 573 |
+
"hash_input_tokens": "742bd6a389a8ef40",
|
| 574 |
+
"hash_cont_tokens": "f11971a765cb609f"
|
| 575 |
+
},
|
| 576 |
+
"truncated": 0,
|
| 577 |
+
"non_truncated": 135,
|
| 578 |
+
"padded": 540,
|
| 579 |
+
"non_padded": 0,
|
| 580 |
+
"effective_few_shots": 5.0,
|
| 581 |
+
"num_truncated_few_shots": 0
|
| 582 |
+
},
|
| 583 |
+
"harness|hendrycksTest-astronomy|5": {
|
| 584 |
+
"hashes": {
|
| 585 |
+
"hash_examples": "7d587b908da4d762",
|
| 586 |
+
"hash_full_prompts": "faf4e80f65de93ca",
|
| 587 |
+
"hash_input_tokens": "aa9743839c83bd9f",
|
| 588 |
+
"hash_cont_tokens": "440a970fadecdc7b"
|
| 589 |
+
},
|
| 590 |
+
"truncated": 0,
|
| 591 |
+
"non_truncated": 152,
|
| 592 |
+
"padded": 608,
|
| 593 |
+
"non_padded": 0,
|
| 594 |
+
"effective_few_shots": 5.0,
|
| 595 |
+
"num_truncated_few_shots": 0
|
| 596 |
+
},
|
| 597 |
+
"harness|hendrycksTest-business_ethics|5": {
|
| 598 |
+
"hashes": {
|
| 599 |
+
"hash_examples": "33e51740670de686",
|
| 600 |
+
"hash_full_prompts": "db01c3ef8e1479d4",
|
| 601 |
+
"hash_input_tokens": "60f6ed52e2a2987a",
|
| 602 |
+
"hash_cont_tokens": "50421e30bef398f9"
|
| 603 |
+
},
|
| 604 |
+
"truncated": 0,
|
| 605 |
+
"non_truncated": 100,
|
| 606 |
+
"padded": 400,
|
| 607 |
+
"non_padded": 0,
|
| 608 |
+
"effective_few_shots": 5.0,
|
| 609 |
+
"num_truncated_few_shots": 0
|
| 610 |
+
},
|
| 611 |
+
"harness|hendrycksTest-clinical_knowledge|5": {
|
| 612 |
+
"hashes": {
|
| 613 |
+
"hash_examples": "f3366dbe7eefffa4",
|
| 614 |
+
"hash_full_prompts": "49654f71d94b65c3",
|
| 615 |
+
"hash_input_tokens": "6080d9f3c5930be0",
|
| 616 |
+
"hash_cont_tokens": "7ecd60c25b9bfe5b"
|
| 617 |
+
},
|
| 618 |
+
"truncated": 0,
|
| 619 |
+
"non_truncated": 265,
|
| 620 |
+
"padded": 1060,
|
| 621 |
+
"non_padded": 0,
|
| 622 |
+
"effective_few_shots": 5.0,
|
| 623 |
+
"num_truncated_few_shots": 0
|
| 624 |
+
},
|
| 625 |
+
"harness|hendrycksTest-college_biology|5": {
|
| 626 |
+
"hashes": {
|
| 627 |
+
"hash_examples": "ca2b6753a0193e7f",
|
| 628 |
+
"hash_full_prompts": "2b460b75f1fdfefd",
|
| 629 |
+
"hash_input_tokens": "873319724ad65589",
|
| 630 |
+
"hash_cont_tokens": "875cde3af7a0ee14"
|
| 631 |
+
},
|
| 632 |
+
"truncated": 0,
|
| 633 |
+
"non_truncated": 144,
|
| 634 |
+
"padded": 564,
|
| 635 |
+
"non_padded": 12,
|
| 636 |
+
"effective_few_shots": 5.0,
|
| 637 |
+
"num_truncated_few_shots": 0
|
| 638 |
+
},
|
| 639 |
+
"harness|hendrycksTest-college_chemistry|5": {
|
| 640 |
+
"hashes": {
|
| 641 |
+
"hash_examples": "22ff85f1d34f42d1",
|
| 642 |
+
"hash_full_prompts": "242c9be6da583e95",
|
| 643 |
+
"hash_input_tokens": "8366d04d12b154a7",
|
| 644 |
+
"hash_cont_tokens": "50421e30bef398f9"
|
| 645 |
+
},
|
| 646 |
+
"truncated": 0,
|
| 647 |
+
"non_truncated": 100,
|
| 648 |
+
"padded": 400,
|
| 649 |
+
"non_padded": 0,
|
| 650 |
+
"effective_few_shots": 5.0,
|
| 651 |
+
"num_truncated_few_shots": 0
|
| 652 |
+
},
|
| 653 |
+
"harness|hendrycksTest-college_computer_science|5": {
|
| 654 |
+
"hashes": {
|
| 655 |
+
"hash_examples": "30318289d717a5cf",
|
| 656 |
+
"hash_full_prompts": "ed2bdb4e87c4b371",
|
| 657 |
+
"hash_input_tokens": "1724a282fb269fd7",
|
| 658 |
+
"hash_cont_tokens": "50421e30bef398f9"
|
| 659 |
+
},
|
| 660 |
+
"truncated": 0,
|
| 661 |
+
"non_truncated": 100,
|
| 662 |
+
"padded": 400,
|
| 663 |
+
"non_padded": 0,
|
| 664 |
+
"effective_few_shots": 5.0,
|
| 665 |
+
"num_truncated_few_shots": 0
|
| 666 |
+
},
|
| 667 |
+
"harness|hendrycksTest-college_mathematics|5": {
|
| 668 |
+
"hashes": {
|
| 669 |
+
"hash_examples": "4944d1f0b6b5d911",
|
| 670 |
+
"hash_full_prompts": "770bc4281c973190",
|
| 671 |
+
"hash_input_tokens": "b7aa815781eae172",
|
| 672 |
+
"hash_cont_tokens": "50421e30bef398f9"
|
| 673 |
+
},
|
| 674 |
+
"truncated": 0,
|
| 675 |
+
"non_truncated": 100,
|
| 676 |
+
"padded": 400,
|
| 677 |
+
"non_padded": 0,
|
| 678 |
+
"effective_few_shots": 5.0,
|
| 679 |
+
"num_truncated_few_shots": 0
|
| 680 |
+
},
|
| 681 |
+
"harness|hendrycksTest-college_medicine|5": {
|
| 682 |
+
"hashes": {
|
| 683 |
+
"hash_examples": "dd69cc33381275af",
|
| 684 |
+
"hash_full_prompts": "ad2a53e5250ab46e",
|
| 685 |
+
"hash_input_tokens": "0003d13e86bc8c1a",
|
| 686 |
+
"hash_cont_tokens": "702fb6d82ff0d6ac"
|
| 687 |
+
},
|
| 688 |
+
"truncated": 0,
|
| 689 |
+
"non_truncated": 173,
|
| 690 |
+
"padded": 692,
|
| 691 |
+
"non_padded": 0,
|
| 692 |
+
"effective_few_shots": 5.0,
|
| 693 |
+
"num_truncated_few_shots": 0
|
| 694 |
+
},
|
| 695 |
+
"harness|hendrycksTest-college_physics|5": {
|
| 696 |
+
"hashes": {
|
| 697 |
+
"hash_examples": "875dd26d22655b0d",
|
| 698 |
+
"hash_full_prompts": "833a0d7b55aed500",
|
| 699 |
+
"hash_input_tokens": "32b28762dd077c78",
|
| 700 |
+
"hash_cont_tokens": "f7b8097afc16a47c"
|
| 701 |
+
},
|
| 702 |
+
"truncated": 0,
|
| 703 |
+
"non_truncated": 102,
|
| 704 |
+
"padded": 404,
|
| 705 |
+
"non_padded": 4,
|
| 706 |
+
"effective_few_shots": 5.0,
|
| 707 |
+
"num_truncated_few_shots": 0
|
| 708 |
+
},
|
| 709 |
+
"harness|hendrycksTest-computer_security|5": {
|
| 710 |
+
"hashes": {
|
| 711 |
+
"hash_examples": "006451eedc0ededb",
|
| 712 |
+
"hash_full_prompts": "94034c97e85d8f46",
|
| 713 |
+
"hash_input_tokens": "19dd0e1895125d49",
|
| 714 |
+
"hash_cont_tokens": "50421e30bef398f9"
|
| 715 |
+
},
|
| 716 |
+
"truncated": 0,
|
| 717 |
+
"non_truncated": 100,
|
| 718 |
+
"padded": 400,
|
| 719 |
+
"non_padded": 0,
|
| 720 |
+
"effective_few_shots": 5.0,
|
| 721 |
+
"num_truncated_few_shots": 0
|
| 722 |
+
},
|
| 723 |
+
"harness|hendrycksTest-conceptual_physics|5": {
|
| 724 |
+
"hashes": {
|
| 725 |
+
"hash_examples": "8874ece872d2ca4c",
|
| 726 |
+
"hash_full_prompts": "e40d15a34640d6fa",
|
| 727 |
+
"hash_input_tokens": "761c7ce187b3338a",
|
| 728 |
+
"hash_cont_tokens": "aa0e8bc655f2f641"
|
| 729 |
+
},
|
| 730 |
+
"truncated": 0,
|
| 731 |
+
"non_truncated": 235,
|
| 732 |
+
"padded": 940,
|
| 733 |
+
"non_padded": 0,
|
| 734 |
+
"effective_few_shots": 5.0,
|
| 735 |
+
"num_truncated_few_shots": 0
|
| 736 |
+
},
|
| 737 |
+
"harness|hendrycksTest-econometrics|5": {
|
| 738 |
+
"hashes": {
|
| 739 |
+
"hash_examples": "64d3623b0bfaa43f",
|
| 740 |
+
"hash_full_prompts": "612f340fae41338d",
|
| 741 |
+
"hash_input_tokens": "dae74024ebc12b2b",
|
| 742 |
+
"hash_cont_tokens": "b1cc6e7e9fcd3827"
|
| 743 |
+
},
|
| 744 |
+
"truncated": 0,
|
| 745 |
+
"non_truncated": 114,
|
| 746 |
+
"padded": 456,
|
| 747 |
+
"non_padded": 0,
|
| 748 |
+
"effective_few_shots": 5.0,
|
| 749 |
+
"num_truncated_few_shots": 0
|
| 750 |
+
},
|
| 751 |
+
"harness|hendrycksTest-electrical_engineering|5": {
|
| 752 |
+
"hashes": {
|
| 753 |
+
"hash_examples": "e98f51780c674d7e",
|
| 754 |
+
"hash_full_prompts": "10275b312d812ae6",
|
| 755 |
+
"hash_input_tokens": "5fa8050688a246ed",
|
| 756 |
+
"hash_cont_tokens": "2425a3f084a591ef"
|
| 757 |
+
},
|
| 758 |
+
"truncated": 0,
|
| 759 |
+
"non_truncated": 145,
|
| 760 |
+
"padded": 580,
|
| 761 |
+
"non_padded": 0,
|
| 762 |
+
"effective_few_shots": 5.0,
|
| 763 |
+
"num_truncated_few_shots": 0
|
| 764 |
+
},
|
| 765 |
+
"harness|hendrycksTest-elementary_mathematics|5": {
|
| 766 |
+
"hashes": {
|
| 767 |
+
"hash_examples": "fc48208a5ac1c0ce",
|
| 768 |
+
"hash_full_prompts": "5ec274c6c82aca23",
|
| 769 |
+
"hash_input_tokens": "2da3f8d7d1515cc6",
|
| 770 |
+
"hash_cont_tokens": "bd87bf0c060fd925"
|
| 771 |
+
},
|
| 772 |
+
"truncated": 0,
|
| 773 |
+
"non_truncated": 378,
|
| 774 |
+
"padded": 1512,
|
| 775 |
+
"non_padded": 0,
|
| 776 |
+
"effective_few_shots": 5.0,
|
| 777 |
+
"num_truncated_few_shots": 0
|
| 778 |
+
},
|
| 779 |
+
"harness|hendrycksTest-formal_logic|5": {
|
| 780 |
+
"hashes": {
|
| 781 |
+
"hash_examples": "5a6525665f63ea72",
|
| 782 |
+
"hash_full_prompts": "07b92638c4a6b500",
|
| 783 |
+
"hash_input_tokens": "907de61bbe46dada",
|
| 784 |
+
"hash_cont_tokens": "eb8932890e0605db"
|
| 785 |
+
},
|
| 786 |
+
"truncated": 0,
|
| 787 |
+
"non_truncated": 126,
|
| 788 |
+
"padded": 504,
|
| 789 |
+
"non_padded": 0,
|
| 790 |
+
"effective_few_shots": 5.0,
|
| 791 |
+
"num_truncated_few_shots": 0
|
| 792 |
+
},
|
| 793 |
+
"harness|hendrycksTest-global_facts|5": {
|
| 794 |
+
"hashes": {
|
| 795 |
+
"hash_examples": "371d70d743b2b89b",
|
| 796 |
+
"hash_full_prompts": "332fdee50a1921b4",
|
| 797 |
+
"hash_input_tokens": "d7549fe9ac133643",
|
| 798 |
+
"hash_cont_tokens": "50421e30bef398f9"
|
| 799 |
+
},
|
| 800 |
+
"truncated": 0,
|
| 801 |
+
"non_truncated": 100,
|
| 802 |
+
"padded": 400,
|
| 803 |
+
"non_padded": 0,
|
| 804 |
+
"effective_few_shots": 5.0,
|
| 805 |
+
"num_truncated_few_shots": 0
|
| 806 |
+
},
|
| 807 |
+
"harness|hendrycksTest-high_school_biology|5": {
|
| 808 |
+
"hashes": {
|
| 809 |
+
"hash_examples": "a79e1018b1674052",
|
| 810 |
+
"hash_full_prompts": "e624e26ede922561",
|
| 811 |
+
"hash_input_tokens": "b449ae8cd622fb96",
|
| 812 |
+
"hash_cont_tokens": "1ddcb86d28cde266"
|
| 813 |
+
},
|
| 814 |
+
"truncated": 0,
|
| 815 |
+
"non_truncated": 310,
|
| 816 |
+
"padded": 1240,
|
| 817 |
+
"non_padded": 0,
|
| 818 |
+
"effective_few_shots": 5.0,
|
| 819 |
+
"num_truncated_few_shots": 0
|
| 820 |
+
},
|
| 821 |
+
"harness|hendrycksTest-high_school_chemistry|5": {
|
| 822 |
+
"hashes": {
|
| 823 |
+
"hash_examples": "44bfc25c389f0e03",
|
| 824 |
+
"hash_full_prompts": "0e3e5f5d9246482a",
|
| 825 |
+
"hash_input_tokens": "a447bd1574b5e26c",
|
| 826 |
+
"hash_cont_tokens": "176c8dcff38c5f8f"
|
| 827 |
+
},
|
| 828 |
+
"truncated": 0,
|
| 829 |
+
"non_truncated": 203,
|
| 830 |
+
"padded": 812,
|
| 831 |
+
"non_padded": 0,
|
| 832 |
+
"effective_few_shots": 5.0,
|
| 833 |
+
"num_truncated_few_shots": 0
|
| 834 |
+
},
|
| 835 |
+
"harness|hendrycksTest-high_school_computer_science|5": {
|
| 836 |
+
"hashes": {
|
| 837 |
+
"hash_examples": "8b8cdb1084f24169",
|
| 838 |
+
"hash_full_prompts": "c00487e67c1813cc",
|
| 839 |
+
"hash_input_tokens": "56312a0c3d85ae90",
|
| 840 |
+
"hash_cont_tokens": "50421e30bef398f9"
|
| 841 |
+
},
|
| 842 |
+
"truncated": 0,
|
| 843 |
+
"non_truncated": 100,
|
| 844 |
+
"padded": 400,
|
| 845 |
+
"non_padded": 0,
|
| 846 |
+
"effective_few_shots": 5.0,
|
| 847 |
+
"num_truncated_few_shots": 0
|
| 848 |
+
},
|
| 849 |
+
"harness|hendrycksTest-high_school_european_history|5": {
|
| 850 |
+
"hashes": {
|
| 851 |
+
"hash_examples": "11cd32d0ef440171",
|
| 852 |
+
"hash_full_prompts": "318f4513c537c6bf",
|
| 853 |
+
"hash_input_tokens": "5002f4ac8b1562ca",
|
| 854 |
+
"hash_cont_tokens": "674fc454bdc5ac93"
|
| 855 |
+
},
|
| 856 |
+
"truncated": 0,
|
| 857 |
+
"non_truncated": 165,
|
| 858 |
+
"padded": 656,
|
| 859 |
+
"non_padded": 4,
|
| 860 |
+
"effective_few_shots": 5.0,
|
| 861 |
+
"num_truncated_few_shots": 0
|
| 862 |
+
},
|
| 863 |
+
"harness|hendrycksTest-high_school_geography|5": {
|
| 864 |
+
"hashes": {
|
| 865 |
+
"hash_examples": "b60019b9e80b642f",
|
| 866 |
+
"hash_full_prompts": "ee5789fcc1a81b1e",
|
| 867 |
+
"hash_input_tokens": "b4f9efd054b0149d",
|
| 868 |
+
"hash_cont_tokens": "03a5012b916274ea"
|
| 869 |
+
},
|
| 870 |
+
"truncated": 0,
|
| 871 |
+
"non_truncated": 198,
|
| 872 |
+
"padded": 792,
|
| 873 |
+
"non_padded": 0,
|
| 874 |
+
"effective_few_shots": 5.0,
|
| 875 |
+
"num_truncated_few_shots": 0
|
| 876 |
+
},
|
| 877 |
+
"harness|hendrycksTest-high_school_government_and_politics|5": {
|
| 878 |
+
"hashes": {
|
| 879 |
+
"hash_examples": "d221ec983d143dc3",
|
| 880 |
+
"hash_full_prompts": "ac42d888e1ce1155",
|
| 881 |
+
"hash_input_tokens": "6e010d01707b5a01",
|
| 882 |
+
"hash_cont_tokens": "873d2aab226ba1d8"
|
| 883 |
+
},
|
| 884 |
+
"truncated": 0,
|
| 885 |
+
"non_truncated": 193,
|
| 886 |
+
"padded": 772,
|
| 887 |
+
"non_padded": 0,
|
| 888 |
+
"effective_few_shots": 5.0,
|
| 889 |
+
"num_truncated_few_shots": 0
|
| 890 |
+
},
|
| 891 |
+
"harness|hendrycksTest-high_school_macroeconomics|5": {
|
| 892 |
+
"hashes": {
|
| 893 |
+
"hash_examples": "59c2915cacfd3fbb",
|
| 894 |
+
"hash_full_prompts": "c6bd9d25158abd0e",
|
| 895 |
+
"hash_input_tokens": "fc1f6e824ba386d7",
|
| 896 |
+
"hash_cont_tokens": "c583432ad27fcfe0"
|
| 897 |
+
},
|
| 898 |
+
"truncated": 0,
|
| 899 |
+
"non_truncated": 390,
|
| 900 |
+
"padded": 1560,
|
| 901 |
+
"non_padded": 0,
|
| 902 |
+
"effective_few_shots": 5.0,
|
| 903 |
+
"num_truncated_few_shots": 0
|
| 904 |
+
},
|
| 905 |
+
"harness|hendrycksTest-high_school_mathematics|5": {
|
| 906 |
+
"hashes": {
|
| 907 |
+
"hash_examples": "1f8ac897608de342",
|
| 908 |
+
"hash_full_prompts": "5d88f41fc2d643a8",
|
| 909 |
+
"hash_input_tokens": "3a485a40c8432ece",
|
| 910 |
+
"hash_cont_tokens": "d7907b61bcb8c123"
|
| 911 |
+
},
|
| 912 |
+
"truncated": 0,
|
| 913 |
+
"non_truncated": 270,
|
| 914 |
+
"padded": 1080,
|
| 915 |
+
"non_padded": 0,
|
| 916 |
+
"effective_few_shots": 5.0,
|
| 917 |
+
"num_truncated_few_shots": 0
|
| 918 |
+
},
|
| 919 |
+
"harness|hendrycksTest-high_school_microeconomics|5": {
|
| 920 |
+
"hashes": {
|
| 921 |
+
"hash_examples": "ead6a0f2f6c83370",
|
| 922 |
+
"hash_full_prompts": "bfc393381298609e",
|
| 923 |
+
"hash_input_tokens": "a7dd9ca4bbda3752",
|
| 924 |
+
"hash_cont_tokens": "f47f041de50333b9"
|
| 925 |
+
},
|
| 926 |
+
"truncated": 0,
|
| 927 |
+
"non_truncated": 238,
|
| 928 |
+
"padded": 952,
|
| 929 |
+
"non_padded": 0,
|
| 930 |
+
"effective_few_shots": 5.0,
|
| 931 |
+
"num_truncated_few_shots": 0
|
| 932 |
+
},
|
| 933 |
+
"harness|hendrycksTest-high_school_physics|5": {
|
| 934 |
+
"hashes": {
|
| 935 |
+
"hash_examples": "c3f2025990afec64",
|
| 936 |
+
"hash_full_prompts": "fc78b4997e436734",
|
| 937 |
+
"hash_input_tokens": "d7ea631399a73865",
|
| 938 |
+
"hash_cont_tokens": "0d56317b3e5eedb5"
|
| 939 |
+
},
|
| 940 |
+
"truncated": 0,
|
| 941 |
+
"non_truncated": 151,
|
| 942 |
+
"padded": 604,
|
| 943 |
+
"non_padded": 0,
|
| 944 |
+
"effective_few_shots": 5.0,
|
| 945 |
+
"num_truncated_few_shots": 0
|
| 946 |
+
},
|
| 947 |
+
"harness|hendrycksTest-high_school_psychology|5": {
|
| 948 |
+
"hashes": {
|
| 949 |
+
"hash_examples": "21f8aab618f6d636",
|
| 950 |
+
"hash_full_prompts": "d5c76aa40b9dbc43",
|
| 951 |
+
"hash_input_tokens": "d12816cf88146011",
|
| 952 |
+
"hash_cont_tokens": "09ba1243e7390c0f"
|
| 953 |
+
},
|
| 954 |
+
"truncated": 0,
|
| 955 |
+
"non_truncated": 545,
|
| 956 |
+
"padded": 2180,
|
| 957 |
+
"non_padded": 0,
|
| 958 |
+
"effective_few_shots": 5.0,
|
| 959 |
+
"num_truncated_few_shots": 0
|
| 960 |
+
},
|
| 961 |
+
"harness|hendrycksTest-high_school_statistics|5": {
|
| 962 |
+
"hashes": {
|
| 963 |
+
"hash_examples": "2386a60a11fc5de3",
|
| 964 |
+
"hash_full_prompts": "4c5c8be5aafac432",
|
| 965 |
+
"hash_input_tokens": "9763ecaef4814c21",
|
| 966 |
+
"hash_cont_tokens": "9cc29889c3d3f77d"
|
| 967 |
+
},
|
| 968 |
+
"truncated": 0,
|
| 969 |
+
"non_truncated": 216,
|
| 970 |
+
"padded": 864,
|
| 971 |
+
"non_padded": 0,
|
| 972 |
+
"effective_few_shots": 5.0,
|
| 973 |
+
"num_truncated_few_shots": 0
|
| 974 |
+
},
|
| 975 |
+
"harness|hendrycksTest-high_school_us_history|5": {
|
| 976 |
+
"hashes": {
|
| 977 |
+
"hash_examples": "74961543be40f04f",
|
| 978 |
+
"hash_full_prompts": "5d5ca4840131ba21",
|
| 979 |
+
"hash_input_tokens": "c639cce12a46ebad",
|
| 980 |
+
"hash_cont_tokens": "cdd0b3dc06d933e5"
|
| 981 |
+
},
|
| 982 |
+
"truncated": 0,
|
| 983 |
+
"non_truncated": 204,
|
| 984 |
+
"padded": 816,
|
| 985 |
+
"non_padded": 0,
|
| 986 |
+
"effective_few_shots": 5.0,
|
| 987 |
+
"num_truncated_few_shots": 0
|
| 988 |
+
},
|
| 989 |
+
"harness|hendrycksTest-high_school_world_history|5": {
|
| 990 |
+
"hashes": {
|
| 991 |
+
"hash_examples": "2ad2f6b7198b2234",
|
| 992 |
+
"hash_full_prompts": "11845057459afd72",
|
| 993 |
+
"hash_input_tokens": "b9762065cce6f3a6",
|
| 994 |
+
"hash_cont_tokens": "e02816433ff28daf"
|
| 995 |
+
},
|
| 996 |
+
"truncated": 0,
|
| 997 |
+
"non_truncated": 237,
|
| 998 |
+
"padded": 948,
|
| 999 |
+
"non_padded": 0,
|
| 1000 |
+
"effective_few_shots": 5.0,
|
| 1001 |
+
"num_truncated_few_shots": 0
|
| 1002 |
+
},
|
| 1003 |
+
"harness|hendrycksTest-human_aging|5": {
|
| 1004 |
+
"hashes": {
|
| 1005 |
+
"hash_examples": "1a7199dc733e779b",
|
| 1006 |
+
"hash_full_prompts": "756b9096b8eaf892",
|
| 1007 |
+
"hash_input_tokens": "84157fee0b6d0f3c",
|
| 1008 |
+
"hash_cont_tokens": "142a4a8a1138a214"
|
| 1009 |
+
},
|
| 1010 |
+
"truncated": 0,
|
| 1011 |
+
"non_truncated": 223,
|
| 1012 |
+
"padded": 892,
|
| 1013 |
+
"non_padded": 0,
|
| 1014 |
+
"effective_few_shots": 5.0,
|
| 1015 |
+
"num_truncated_few_shots": 0
|
| 1016 |
+
},
|
| 1017 |
+
"harness|hendrycksTest-human_sexuality|5": {
|
| 1018 |
+
"hashes": {
|
| 1019 |
+
"hash_examples": "7acb8fdad97f88a6",
|
| 1020 |
+
"hash_full_prompts": "731a52ff15b8cfdb",
|
| 1021 |
+
"hash_input_tokens": "ade303e1ae3c016f",
|
| 1022 |
+
"hash_cont_tokens": "bc54813e809b796d"
|
| 1023 |
+
},
|
| 1024 |
+
"truncated": 0,
|
| 1025 |
+
"non_truncated": 131,
|
| 1026 |
+
"padded": 524,
|
| 1027 |
+
"non_padded": 0,
|
| 1028 |
+
"effective_few_shots": 5.0,
|
| 1029 |
+
"num_truncated_few_shots": 0
|
| 1030 |
+
},
|
| 1031 |
+
"harness|hendrycksTest-international_law|5": {
|
| 1032 |
+
"hashes": {
|
| 1033 |
+
"hash_examples": "1300bfd0dfc59114",
|
| 1034 |
+
"hash_full_prompts": "db2aefbff5eec996",
|
| 1035 |
+
"hash_input_tokens": "e5482e1c23c23d35",
|
| 1036 |
+
"hash_cont_tokens": "8ea8c5ff76a15bca"
|
| 1037 |
+
},
|
| 1038 |
+
"truncated": 0,
|
| 1039 |
+
"non_truncated": 121,
|
| 1040 |
+
"padded": 484,
|
| 1041 |
+
"non_padded": 0,
|
| 1042 |
+
"effective_few_shots": 5.0,
|
| 1043 |
+
"num_truncated_few_shots": 0
|
| 1044 |
+
},
|
| 1045 |
+
"harness|hendrycksTest-jurisprudence|5": {
|
| 1046 |
+
"hashes": {
|
| 1047 |
+
"hash_examples": "083b1e4904c48dc2",
|
| 1048 |
+
"hash_full_prompts": "0f89ee3fe03d6a21",
|
| 1049 |
+
"hash_input_tokens": "4415eeb9bad0507b",
|
| 1050 |
+
"hash_cont_tokens": "e3a8cd951b6e3469"
|
| 1051 |
+
},
|
| 1052 |
+
"truncated": 0,
|
| 1053 |
+
"non_truncated": 108,
|
| 1054 |
+
"padded": 432,
|
| 1055 |
+
"non_padded": 0,
|
| 1056 |
+
"effective_few_shots": 5.0,
|
| 1057 |
+
"num_truncated_few_shots": 0
|
| 1058 |
+
},
|
| 1059 |
+
"harness|hendrycksTest-logical_fallacies|5": {
|
| 1060 |
+
"hashes": {
|
| 1061 |
+
"hash_examples": "709128f9926a634c",
|
| 1062 |
+
"hash_full_prompts": "98a04b1f8f841069",
|
| 1063 |
+
"hash_input_tokens": "e6b5271422ecbaa8",
|
| 1064 |
+
"hash_cont_tokens": "3e9e0bdc248fd88a"
|
| 1065 |
+
},
|
| 1066 |
+
"truncated": 0,
|
| 1067 |
+
"non_truncated": 163,
|
| 1068 |
+
"padded": 644,
|
| 1069 |
+
"non_padded": 8,
|
| 1070 |
+
"effective_few_shots": 5.0,
|
| 1071 |
+
"num_truncated_few_shots": 0
|
| 1072 |
+
},
|
| 1073 |
+
"harness|hendrycksTest-machine_learning|5": {
|
| 1074 |
+
"hashes": {
|
| 1075 |
+
"hash_examples": "88f22a636029ae47",
|
| 1076 |
+
"hash_full_prompts": "2e1c8d4b1e0cc921",
|
| 1077 |
+
"hash_input_tokens": "e719cb83196977d8",
|
| 1078 |
+
"hash_cont_tokens": "55b12fb138c6a064"
|
| 1079 |
+
},
|
| 1080 |
+
"truncated": 0,
|
| 1081 |
+
"non_truncated": 112,
|
| 1082 |
+
"padded": 448,
|
| 1083 |
+
"non_padded": 0,
|
| 1084 |
+
"effective_few_shots": 5.0,
|
| 1085 |
+
"num_truncated_few_shots": 0
|
| 1086 |
+
},
|
| 1087 |
+
"harness|hendrycksTest-management|5": {
|
| 1088 |
+
"hashes": {
|
| 1089 |
+
"hash_examples": "8c8a1e07a2151dca",
|
| 1090 |
+
"hash_full_prompts": "f51611f514b265b0",
|
| 1091 |
+
"hash_input_tokens": "155da0e62b39e804",
|
| 1092 |
+
"hash_cont_tokens": "a01d6d39a83c4597"
|
| 1093 |
+
},
|
| 1094 |
+
"truncated": 0,
|
| 1095 |
+
"non_truncated": 103,
|
| 1096 |
+
"padded": 412,
|
| 1097 |
+
"non_padded": 0,
|
| 1098 |
+
"effective_few_shots": 5.0,
|
| 1099 |
+
"num_truncated_few_shots": 0
|
| 1100 |
+
},
|
| 1101 |
+
"harness|hendrycksTest-marketing|5": {
|
| 1102 |
+
"hashes": {
|
| 1103 |
+
"hash_examples": "2668953431f91e96",
|
| 1104 |
+
"hash_full_prompts": "77562bef997c7650",
|
| 1105 |
+
"hash_input_tokens": "38466c242259e6d3",
|
| 1106 |
+
"hash_cont_tokens": "6aeaed4d823c98aa"
|
| 1107 |
+
},
|
| 1108 |
+
"truncated": 0,
|
| 1109 |
+
"non_truncated": 234,
|
| 1110 |
+
"padded": 932,
|
| 1111 |
+
"non_padded": 4,
|
| 1112 |
+
"effective_few_shots": 5.0,
|
| 1113 |
+
"num_truncated_few_shots": 0
|
| 1114 |
+
},
|
| 1115 |
+
"harness|hendrycksTest-medical_genetics|5": {
|
| 1116 |
+
"hashes": {
|
| 1117 |
+
"hash_examples": "9c2dda34a2ea4fd2",
|
| 1118 |
+
"hash_full_prompts": "202139046daa118f",
|
| 1119 |
+
"hash_input_tokens": "0dd129e92538a7f6",
|
| 1120 |
+
"hash_cont_tokens": "50421e30bef398f9"
|
| 1121 |
+
},
|
| 1122 |
+
"truncated": 0,
|
| 1123 |
+
"non_truncated": 100,
|
| 1124 |
+
"padded": 400,
|
| 1125 |
+
"non_padded": 0,
|
| 1126 |
+
"effective_few_shots": 5.0,
|
| 1127 |
+
"num_truncated_few_shots": 0
|
| 1128 |
+
},
|
| 1129 |
+
"harness|hendrycksTest-miscellaneous|5": {
|
| 1130 |
+
"hashes": {
|
| 1131 |
+
"hash_examples": "41adb694024809c2",
|
| 1132 |
+
"hash_full_prompts": "bffec9fc237bcf93",
|
| 1133 |
+
"hash_input_tokens": "d108a883fc3e022f",
|
| 1134 |
+
"hash_cont_tokens": "9b0ab02a64603081"
|
| 1135 |
+
},
|
| 1136 |
+
"truncated": 0,
|
| 1137 |
+
"non_truncated": 783,
|
| 1138 |
+
"padded": 3132,
|
| 1139 |
+
"non_padded": 0,
|
| 1140 |
+
"effective_few_shots": 5.0,
|
| 1141 |
+
"num_truncated_few_shots": 0
|
| 1142 |
+
},
|
| 1143 |
+
"harness|hendrycksTest-moral_disputes|5": {
|
| 1144 |
+
"hashes": {
|
| 1145 |
+
"hash_examples": "3171c13ba3c594c4",
|
| 1146 |
+
"hash_full_prompts": "170831fc36f1d59e",
|
| 1147 |
+
"hash_input_tokens": "0e7b7df82884a2d5",
|
| 1148 |
+
"hash_cont_tokens": "3b8bbe9108e55ce9"
|
| 1149 |
+
},
|
| 1150 |
+
"truncated": 0,
|
| 1151 |
+
"non_truncated": 346,
|
| 1152 |
+
"padded": 1364,
|
| 1153 |
+
"non_padded": 20,
|
| 1154 |
+
"effective_few_shots": 5.0,
|
| 1155 |
+
"num_truncated_few_shots": 0
|
| 1156 |
+
},
|
| 1157 |
+
"harness|hendrycksTest-moral_scenarios|5": {
|
| 1158 |
+
"hashes": {
|
| 1159 |
+
"hash_examples": "9873e077e83e0546",
|
| 1160 |
+
"hash_full_prompts": "08f4ceba3131a068",
|
| 1161 |
+
"hash_input_tokens": "7c220f5613cd8426",
|
| 1162 |
+
"hash_cont_tokens": "3e9bfc0362e97330"
|
| 1163 |
+
},
|
| 1164 |
+
"truncated": 0,
|
| 1165 |
+
"non_truncated": 895,
|
| 1166 |
+
"padded": 3580,
|
| 1167 |
+
"non_padded": 0,
|
| 1168 |
+
"effective_few_shots": 5.0,
|
| 1169 |
+
"num_truncated_few_shots": 0
|
| 1170 |
+
},
|
| 1171 |
+
"harness|hendrycksTest-nutrition|5": {
|
| 1172 |
+
"hashes": {
|
| 1173 |
+
"hash_examples": "7db1d8142ec14323",
|
| 1174 |
+
"hash_full_prompts": "4c0e68e3586cb453",
|
| 1175 |
+
"hash_input_tokens": "35de1609a9a763a9",
|
| 1176 |
+
"hash_cont_tokens": "23b2dc6ee2da4cfc"
|
| 1177 |
+
},
|
| 1178 |
+
"truncated": 0,
|
| 1179 |
+
"non_truncated": 306,
|
| 1180 |
+
"padded": 1224,
|
| 1181 |
+
"non_padded": 0,
|
| 1182 |
+
"effective_few_shots": 5.0,
|
| 1183 |
+
"num_truncated_few_shots": 0
|
| 1184 |
+
},
|
| 1185 |
+
"harness|hendrycksTest-philosophy|5": {
|
| 1186 |
+
"hashes": {
|
| 1187 |
+
"hash_examples": "9b455b7d72811cc8",
|
| 1188 |
+
"hash_full_prompts": "e467f822d8a0d3ff",
|
| 1189 |
+
"hash_input_tokens": "a1dcfa9c80490d06",
|
| 1190 |
+
"hash_cont_tokens": "9f6ff69d23a48783"
|
| 1191 |
+
},
|
| 1192 |
+
"truncated": 0,
|
| 1193 |
+
"non_truncated": 311,
|
| 1194 |
+
"padded": 1244,
|
| 1195 |
+
"non_padded": 0,
|
| 1196 |
+
"effective_few_shots": 5.0,
|
| 1197 |
+
"num_truncated_few_shots": 0
|
| 1198 |
+
},
|
| 1199 |
+
"harness|hendrycksTest-prehistory|5": {
|
| 1200 |
+
"hashes": {
|
| 1201 |
+
"hash_examples": "8be90d0f538f1560",
|
| 1202 |
+
"hash_full_prompts": "152187949bcd0921",
|
| 1203 |
+
"hash_input_tokens": "a091cf645d2415e0",
|
| 1204 |
+
"hash_cont_tokens": "d6458d743d875837"
|
| 1205 |
+
},
|
| 1206 |
+
"truncated": 0,
|
| 1207 |
+
"non_truncated": 324,
|
| 1208 |
+
"padded": 1296,
|
| 1209 |
+
"non_padded": 0,
|
| 1210 |
+
"effective_few_shots": 5.0,
|
| 1211 |
+
"num_truncated_few_shots": 0
|
| 1212 |
+
},
|
| 1213 |
+
"harness|hendrycksTest-professional_accounting|5": {
|
| 1214 |
+
"hashes": {
|
| 1215 |
+
"hash_examples": "8d377597916cd07e",
|
| 1216 |
+
"hash_full_prompts": "0eb7345d6144ee0d",
|
| 1217 |
+
"hash_input_tokens": "e9df32a33f85290c",
|
| 1218 |
+
"hash_cont_tokens": "922a195f53a35662"
|
| 1219 |
+
},
|
| 1220 |
+
"truncated": 0,
|
| 1221 |
+
"non_truncated": 282,
|
| 1222 |
+
"padded": 1128,
|
| 1223 |
+
"non_padded": 0,
|
| 1224 |
+
"effective_few_shots": 5.0,
|
| 1225 |
+
"num_truncated_few_shots": 0
|
| 1226 |
+
},
|
| 1227 |
+
"harness|hendrycksTest-professional_law|5": {
|
| 1228 |
+
"hashes": {
|
| 1229 |
+
"hash_examples": "cd9dbc52b3c932d6",
|
| 1230 |
+
"hash_full_prompts": "36ac764272bfb182",
|
| 1231 |
+
"hash_input_tokens": "c9f7583fff66d361",
|
| 1232 |
+
"hash_cont_tokens": "2e590029ef41fbcd"
|
| 1233 |
+
},
|
| 1234 |
+
"truncated": 0,
|
| 1235 |
+
"non_truncated": 1534,
|
| 1236 |
+
"padded": 6136,
|
| 1237 |
+
"non_padded": 0,
|
| 1238 |
+
"effective_few_shots": 5.0,
|
| 1239 |
+
"num_truncated_few_shots": 0
|
| 1240 |
+
},
|
| 1241 |
+
"harness|hendrycksTest-professional_medicine|5": {
|
| 1242 |
+
"hashes": {
|
| 1243 |
+
"hash_examples": "b20e4e816c1e383e",
|
| 1244 |
+
"hash_full_prompts": "7b8d69ea2acaf2f7",
|
| 1245 |
+
"hash_input_tokens": "40a933f829116f8d",
|
| 1246 |
+
"hash_cont_tokens": "7cfee54dbddd5a98"
|
| 1247 |
+
},
|
| 1248 |
+
"truncated": 0,
|
| 1249 |
+
"non_truncated": 272,
|
| 1250 |
+
"padded": 1088,
|
| 1251 |
+
"non_padded": 0,
|
| 1252 |
+
"effective_few_shots": 5.0,
|
| 1253 |
+
"num_truncated_few_shots": 0
|
| 1254 |
+
},
|
| 1255 |
+
"harness|hendrycksTest-professional_psychology|5": {
|
| 1256 |
+
"hashes": {
|
| 1257 |
+
"hash_examples": "d45b73b22f9cc039",
|
| 1258 |
+
"hash_full_prompts": "fe8937e9ffc99771",
|
| 1259 |
+
"hash_input_tokens": "0f6a92c3a2062b48",
|
| 1260 |
+
"hash_cont_tokens": "a86677b2a45c20e1"
|
| 1261 |
+
},
|
| 1262 |
+
"truncated": 0,
|
| 1263 |
+
"non_truncated": 612,
|
| 1264 |
+
"padded": 2448,
|
| 1265 |
+
"non_padded": 0,
|
| 1266 |
+
"effective_few_shots": 5.0,
|
| 1267 |
+
"num_truncated_few_shots": 0
|
| 1268 |
+
},
|
| 1269 |
+
"harness|hendrycksTest-public_relations|5": {
|
| 1270 |
+
"hashes": {
|
| 1271 |
+
"hash_examples": "0d25072e1761652a",
|
| 1272 |
+
"hash_full_prompts": "f9adc39cfa9f42ba",
|
| 1273 |
+
"hash_input_tokens": "29a08e9bfbe9b2f0",
|
| 1274 |
+
"hash_cont_tokens": "0d756ccaae031757"
|
| 1275 |
+
},
|
| 1276 |
+
"truncated": 0,
|
| 1277 |
+
"non_truncated": 110,
|
| 1278 |
+
"padded": 440,
|
| 1279 |
+
"non_padded": 0,
|
| 1280 |
+
"effective_few_shots": 5.0,
|
| 1281 |
+
"num_truncated_few_shots": 0
|
| 1282 |
+
},
|
| 1283 |
+
"harness|hendrycksTest-security_studies|5": {
|
| 1284 |
+
"hashes": {
|
| 1285 |
+
"hash_examples": "62bb8197e63d60d4",
|
| 1286 |
+
"hash_full_prompts": "869c9c3ae196b7c3",
|
| 1287 |
+
"hash_input_tokens": "32a03f1f22a6e103",
|
| 1288 |
+
"hash_cont_tokens": "b2229bc2cfbf594b"
|
| 1289 |
+
},
|
| 1290 |
+
"truncated": 0,
|
| 1291 |
+
"non_truncated": 245,
|
| 1292 |
+
"padded": 980,
|
| 1293 |
+
"non_padded": 0,
|
| 1294 |
+
"effective_few_shots": 5.0,
|
| 1295 |
+
"num_truncated_few_shots": 0
|
| 1296 |
+
},
|
| 1297 |
+
"harness|hendrycksTest-sociology|5": {
|
| 1298 |
+
"hashes": {
|
| 1299 |
+
"hash_examples": "e7959df87dea8672",
|
| 1300 |
+
"hash_full_prompts": "1a1fc00e17b3a52a",
|
| 1301 |
+
"hash_input_tokens": "1de5c52d2b2831d7",
|
| 1302 |
+
"hash_cont_tokens": "c3a3bdfd177eed5b"
|
| 1303 |
+
},
|
| 1304 |
+
"truncated": 0,
|
| 1305 |
+
"non_truncated": 201,
|
| 1306 |
+
"padded": 800,
|
| 1307 |
+
"non_padded": 4,
|
| 1308 |
+
"effective_few_shots": 5.0,
|
| 1309 |
+
"num_truncated_few_shots": 0
|
| 1310 |
+
},
|
| 1311 |
+
"harness|hendrycksTest-us_foreign_policy|5": {
|
| 1312 |
+
"hashes": {
|
| 1313 |
+
"hash_examples": "4a56a01ddca44dca",
|
| 1314 |
+
"hash_full_prompts": "0c7a7081c71c07b6",
|
| 1315 |
+
"hash_input_tokens": "add924961f7f4146",
|
| 1316 |
+
"hash_cont_tokens": "50421e30bef398f9"
|
| 1317 |
+
},
|
| 1318 |
+
"truncated": 0,
|
| 1319 |
+
"non_truncated": 100,
|
| 1320 |
+
"padded": 400,
|
| 1321 |
+
"non_padded": 0,
|
| 1322 |
+
"effective_few_shots": 5.0,
|
| 1323 |
+
"num_truncated_few_shots": 0
|
| 1324 |
+
},
|
| 1325 |
+
"harness|hendrycksTest-virology|5": {
|
| 1326 |
+
"hashes": {
|
| 1327 |
+
"hash_examples": "451cc86a8c4f4fe9",
|
| 1328 |
+
"hash_full_prompts": "01e95325d8b738e4",
|
| 1329 |
+
"hash_input_tokens": "e0653601c466b1bc",
|
| 1330 |
+
"hash_cont_tokens": "af8b3658088cb37f"
|
| 1331 |
+
},
|
| 1332 |
+
"truncated": 0,
|
| 1333 |
+
"non_truncated": 166,
|
| 1334 |
+
"padded": 664,
|
| 1335 |
+
"non_padded": 0,
|
| 1336 |
+
"effective_few_shots": 5.0,
|
| 1337 |
+
"num_truncated_few_shots": 0
|
| 1338 |
+
},
|
| 1339 |
+
"harness|hendrycksTest-world_religions|5": {
|
| 1340 |
+
"hashes": {
|
| 1341 |
+
"hash_examples": "3b29cfaf1a81c379",
|
| 1342 |
+
"hash_full_prompts": "e0d79a15083dfdff",
|
| 1343 |
+
"hash_input_tokens": "ac600d612445156d",
|
| 1344 |
+
"hash_cont_tokens": "060118bef6de4e0a"
|
| 1345 |
+
},
|
| 1346 |
+
"truncated": 0,
|
| 1347 |
+
"non_truncated": 171,
|
| 1348 |
+
"padded": 684,
|
| 1349 |
+
"non_padded": 0,
|
| 1350 |
+
"effective_few_shots": 5.0,
|
| 1351 |
+
"num_truncated_few_shots": 0
|
| 1352 |
+
},
|
| 1353 |
+
"harness|truthfulqa:mc|0": {
|
| 1354 |
+
"hashes": {
|
| 1355 |
+
"hash_examples": "23176c0531c7b867",
|
| 1356 |
+
"hash_full_prompts": "36a6d90e75d92d4a",
|
| 1357 |
+
"hash_input_tokens": "a03ce28b7fd06aa7",
|
| 1358 |
+
"hash_cont_tokens": "f5da56a132aab151"
|
| 1359 |
+
},
|
| 1360 |
+
"truncated": 0,
|
| 1361 |
+
"non_truncated": 817,
|
| 1362 |
+
"padded": 9996,
|
| 1363 |
+
"non_padded": 0,
|
| 1364 |
+
"effective_few_shots": 0.0,
|
| 1365 |
+
"num_truncated_few_shots": 0
|
| 1366 |
+
},
|
| 1367 |
+
"harness|winogrande|5": {
|
| 1368 |
+
"hashes": {
|
| 1369 |
+
"hash_examples": "aada0a176fd81218",
|
| 1370 |
+
"hash_full_prompts": "c8655cbd12de8409",
|
| 1371 |
+
"hash_input_tokens": "72067255e368e24e",
|
| 1372 |
+
"hash_cont_tokens": "f08975ad6f2d5864"
|
| 1373 |
+
},
|
| 1374 |
+
"truncated": 0,
|
| 1375 |
+
"non_truncated": 1267,
|
| 1376 |
+
"padded": 2534,
|
| 1377 |
+
"non_padded": 0,
|
| 1378 |
+
"effective_few_shots": 5.0,
|
| 1379 |
+
"num_truncated_few_shots": 0
|
| 1380 |
+
},
|
| 1381 |
+
"harness|gsm8k|5": {
|
| 1382 |
+
"hashes": {
|
| 1383 |
+
"hash_examples": "4c0843a5d99bcfdc",
|
| 1384 |
+
"hash_full_prompts": "41d55e83abc0e02d",
|
| 1385 |
+
"hash_input_tokens": "bda342e47b5099b2",
|
| 1386 |
+
"hash_cont_tokens": "e4101d08d98273ca"
|
| 1387 |
+
},
|
| 1388 |
+
"truncated": 0,
|
| 1389 |
+
"non_truncated": 1319,
|
| 1390 |
+
"padded": 0,
|
| 1391 |
+
"non_padded": 1319,
|
| 1392 |
+
"effective_few_shots": 5.0,
|
| 1393 |
+
"num_truncated_few_shots": 0
|
| 1394 |
+
}
|
| 1395 |
+
},
|
| 1396 |
+
"summary_general": {
|
| 1397 |
+
"hashes": {
|
| 1398 |
+
"hash_examples": "3b7fa57a057f9415",
|
| 1399 |
+
"hash_full_prompts": "63615fc50fc9417c",
|
| 1400 |
+
"hash_input_tokens": "a8fa53915153e1db",
|
| 1401 |
+
"hash_cont_tokens": "c3c012687e8b60d2"
|
| 1402 |
+
},
|
| 1403 |
+
"truncated": 0,
|
| 1404 |
+
"non_truncated": 28659,
|
| 1405 |
+
"padded": 113348,
|
| 1406 |
+
"non_padded": 1524,
|
| 1407 |
+
"num_truncated_few_shots": 0
|
| 1408 |
+
}
|
| 1409 |
+
}
|
results_2024-05-27T06-04-08.229332.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0daf00c9bab17dedff430aa060f6a247564190f4471e3222f4065e643e7d516c
|
| 3 |
+
size 708892
|