Add some examples to the synthetic ds generation prompt

generate_synthetic_dataset.py

@@ -15,6 +15,35 @@ client = GrazieApiGatewayClient(
 )
 
 
+def get_example_prompt(start_msg, end_msg):
+    return f"""START OF THE EXAMPLE
+
+For following the edited message:
+START OF THE EDITED COMMIT MESSAGE 
+{end_msg}
+END OF THE EDITED COMMIT MESSAGE
+
+You would output the following initial commit message:
+START OF THE INITIAL COMMIT MESSAGE
+{start_msg}
+END OF THE INITIAL COMMIT MESSAGE
+
+END OF THE EXAMPLE"""
+
+
+def generate_examples():
+    manual_df = hf_data_loader.load_raw_rewriting_dataset_as_pandas()[['commit_msg_start', 'commit_msg_end']]
+    examples = [
+        get_example_prompt(row['commit_msg_start'], row['commit_msg_end'])
+        for _, row in manual_df.iterrows()
+    ]
+
+    return "\n".join(examples)
+
+
+EXAMPLES = generate_examples()
+
+
 def build_prompt(reference, diff):
     return f"""A software developer uses a LLM to generate commit messages.
     
@@ -29,7 +58,12 @@ START OF THE COMMIT MESSAGE
 {reference}
 END OF THE COMMIT MESSAGE
 
-Your task is to print the initial, LLM-generated commit message. Print only the initial commit message's text after the 
+Your task is to print the initial, LLM-generated commit message. Here are some examples of what you should output:
+START OF THE EXAMPLES LIST
+{EXAMPLES}
+END OF THE EXAMPLES LIST
+
+Print only the initial commit message's text after the 
 token "OUTPUT".
 
 OUTPUT"""