-
pass@1:
- Only one generated answer is allowed for each question.
- The model passes if the single generated answer is correct.
-
match@8:
- Eight answers are sampled.
- The model passes if the majority (at least 5 out of 8) are correct.
-
pass@8:
- Eight answers are sampled.
- The model passes if at least one of the eight generated answers is correct.
cualignment/modeldiffing
Folders and files
| Name | Name | Last commit date | ||
|---|---|---|---|---|