IBM / unitxt / 19036532707

81%

push

github

Correct reflection based tool calling metrics so valid results will be 1. (#1940)

* Correct reflection based tool calling metrics so valid results will be 1.

Also added example of using reflection based tool calling to correct tool calls and reevaluate.

Signed-off-by: Yoav Katz <katz@il.ibm.com>

* Fixed the test of the tool calling evaluation metrics accordingly.

---------

Signed-off-by: Yoav Katz <katz@il.ibm.com>
Co-authored-by: Koren Lazar <koren.lazar@ibm.com>

1607 of 2006 branches covered (80.11%)

Branch coverage included in aggregate %.

10947 of 13514 relevant lines covered (81.0%)

0.81 hits per line