31 lines
1.1 KiB
TeX
31 lines
1.1 KiB
TeX
\subsection{Evaluation Workflow Pseudocode}
|
|
|
|
\begin{algorithm}[t]
|
|
\caption{OPT--Code Evaluation and Adjudication Pipeline}
|
|
\label{alg:opt-eval-pipeline}
|
|
\begin{algorithmic}[1]
|
|
\Require System description $S$
|
|
\State $C_A \gets$ ClassifierModel($S$, MinimalPrompt or MaximalPrompt)
|
|
\State $C_B \gets$ ClassifierModel($S$, MinimalPrompt or MaximalPrompt)
|
|
\State $E_A \gets$ EvaluatorModel($S, C_A$, EvaluatorPrompt)
|
|
\State $E_B \gets$ EvaluatorModel($S, C_B$, EvaluatorPrompt)
|
|
|
|
\If{($E_A.\text{verdict}$ and $E_B.\text{verdict}$) are acceptable}
|
|
\If{$C_A.\text{OPT} = C_B.\text{OPT}$}
|
|
\State \Return $C_A$ as final OPT--Code
|
|
\Else
|
|
\State $J \gets$ AdjudicatorModel($S, C_A, E_A, C_B, E_B$, AdjudicatorPrompt)
|
|
\State $C^\ast \gets J.\text{Final OPT--Code}$
|
|
\State $E^\ast \gets$ EvaluatorModel($S, C^\ast$, EvaluatorPrompt)
|
|
\If{$E^\ast.\text{verdict}$ acceptable}
|
|
\State \Return $C^\ast$ as final OPT--Code
|
|
\Else
|
|
\State Flag case for human review
|
|
\EndIf
|
|
\EndIf
|
|
\Else
|
|
\State Flag case for human review
|
|
\EndIf
|
|
\end{algorithmic}
|
|
\end{algorithm}
|