Change ret value from single dict to multiple ret values

Signed-off-by: Nathan Weinberg <[email protected]>

src/instructlab/eval/mmlu.py

@@ -25,11 +25,7 @@ def __init__(
     def run(self) -> dict:
         individual_scores: dict[str, float] = {}
         overall_score: float = 0.0
-        payload = {
-            "individual_scores": individual_scores,
-            "overall_score": overall_score,
-        }
-        return payload
+        return overall_score, individual_scores
 
 
 class PR_MMLU_Evaluator(Evaluator):
@@ -39,8 +35,8 @@ class PR_MMLU_Evaluator(Evaluator):
     Attributes:
         sdg_path    path where all the PR MMLU tasks are stored
         task        group name that is shared by all the PR MMLU tasks
-        few_shots    number of examples
-        batch_size   number of GPUs
+        few_shots   number of examples
+        batch_size  number of GPUs
     """
 
     def __init__(
@@ -61,9 +57,4 @@ def run(self) -> dict:
         individual_scores: dict[str, float] = {}
         overall_score: float = 0.0
         qa_pairs: list[tuple] = []
-        payload = {
-            "individual_scores": individual_scores,
-            "overall_score": overall_score,
-            "qa_pairs": qa_pairs,
-        }
-        return payload
+        return overall_score, individual_scores, qa_pairs

src/instructlab/eval/mtbench.py

@@ -19,8 +19,7 @@ def __init__(self, model_path, server_url: str) -> None:
     def run(self) -> dict:
         overall_score: float = 0.0
         qa_pairs: list[tuple] = []
-        payload = {"overall_score": overall_score, "qa_pairs": qa_pairs}
-        return payload
+        return overall_score, qa_pairs
 
 
 class PR_Bench_Evaluator(Evaluator):
@@ -40,5 +39,4 @@ def __init__(self, model_path, server_url: str, questions: str) -> None:
     def run(self) -> dict:
         overall_score = 0.0
         qa_pairs: list[tuple] = []
-        payload = {"overall_score": overall_score, "qa_pairs": qa_pairs}
-        return payload
+        return overall_score, qa_pairs