IBM / unitxt / 15581348656 / 1

81%

main: 81%

push

github

Bluebench fixes (#1828)

* Fix Arena Hard template to use latest llama judge model (used in Bluebench)

Signed-off-by: Jonathan Bnayahu <bnayahu@il.ibm.com>

* Update the Arena Hard recipe in Bluebench to use llama-3-3-70b-instruct as judge.

Signed-off-by: Jonathan Bnayahu <bnayahu@il.ibm.com>

* Add a requirements section for bluebench

Signed-off-by: Jonathan Bnayahu <bnayahu@il.ibm.com>

---------

Signed-off-by: Jonathan Bnayahu <bnayahu@il.ibm.com>

1689 of 2081 branches covered (81.16%)

Branch coverage included in aggregate %.

10478 of 13082 relevant lines covered (80.09%)

0.8 hits per line