added fix split, gitignore and download mmlu script

end-to-end-use-cases/prompt-migration/.gitignore

@@ -0,0 +1 @@
+/mmlu_pro_data

end-to-end-use-cases/prompt-migration/benchmarks/download_mmlu_pro.py

@@ -0,0 +1,27 @@
+from datasets import load_dataset
+import pandas as pd
+import os
+
+def download_mmlu_pro():
+    # Create output directory if it doesn't exist
+    output_dir = "mmlu_pro_data"
+    os.makedirs(output_dir, exist_ok=True)
+    
+    # Load the dataset
+    dataset = load_dataset("TIGER-Lab/MMLU-Pro")
+    
+    # Convert each split to CSV
+    for split in dataset.keys():
+        # Convert to pandas DataFrame
+        df = pd.DataFrame(dataset[split])
+        
+        # Save to CSV
+        output_path = os.path.join(output_dir, f"mmlu_pro_{split}.csv")
+        df.to_csv(output_path, index=False)
+        print(f"Saved {split} split to {output_path}")
+        print(f"Number of examples in {split}: {len(df)}")
+        
+if __name__ == "__main__":
+    print("Downloading MMLU-Pro dataset...")
+    download_mmlu_pro()
+    print("Download complete!")

end-to-end-use-cases/prompt-migration/benchmarks/helpers.py

@@ -23,3 +23,31 @@ def train_val_test_split(
         valset=list(map(mapper, validation_docs)),
         testset=list(map(mapper, test_docs)),
     )
+
+
+def fixed_split(
+    dataset: "Dataset",
+    mapper: t.Callable[[dict], "dspy.Example"],
+    train_size: int = 1000,
+    validation_size: int = 200,
+) -> TaskDatasets:
+    """Split dataset by taking first N examples instead of random sampling.
+    
+    Args:
+        dataset: Input dataset
+        mapper: Function to map dataset examples to dspy.Example
+        train_size: Number of examples to use for training (default: 1000)
+        validation_size: Number of examples to use for validation (default: 200)
+    
+    Returns:
+        TaskDatasets containing train, validation and test splits
+    """
+    train_docs = dataset.select(range(train_size))
+    validation_docs = dataset.select(range(train_size, train_size + validation_size))
+    test_docs = dataset.select(range(train_size + validation_size, len(dataset)))
+    
+    return TaskDatasets(
+        trainset=list(map(mapper, train_docs)),
+        valset=list(map(mapper, validation_docs)),
+        testset=list(map(mapper, test_docs)),
+    )

end-to-end-use-cases/prompt-migration/benchmarks/llama_mmlu_pro.py

@@ -55,10 +55,11 @@ def _task_doc_example(doc: TaskDoc) -> dspy.Example:
     example = dspy.Example(
         question=doc["input_question"],
         options=doc["input_choice_list"],
+        reasoning="",
         answer=doc["output_parsed_answer"],
     )
     example._input_keys = {"question", "options"}
-    example._output_keys = {"answer"}
+    example._output_keys = {"reasoning", "answer"} 
     return example