Get better visuals for Prometheus Eval

evaluation/prometheus_evaluation_analysis.py

@@ -3,10 +3,10 @@ import matplotlib.pyplot as plt
 import glob
 
 # Adjust the file pattern/path as needed
-csv_files = sorted(glob.glob("eval_reports/qwen-base-responses-evaluation-pass*.csv"))[:10]
+csv_files = sorted(glob.glob("eval_reports/qwen-finetuned-responses-evaluation-pass*.csv"))[:10]
 print(csv_files)
 # Create a subplot grid: 2 rows x 5 columns for 10 plots
-fig, axes = plt.subplots(nrows=5, ncols=2, figsize=(10, 30))
+fig, axes = plt.subplots(nrows=5, ncols=2, figsize=(30, 30))
 axes = axes.flatten()
 
 for i, csv_file in enumerate(csv_files):
@@ -34,11 +34,12 @@ for i, csv_file in enumerate(csv_files):
     ax = axes[i]
     ax.bar(list(score_counts.keys()), list(score_counts.values()),
            color='skyblue', edgecolor='black')
-    ax.set_title(f"{csv_file}\nMean: {mean_score:.2f}, Mode: {mode_str}")
-    ax.set_xlabel("Score")
-    ax.set_ylabel("Frequency")
-    ax.tick_params(axis='x', rotation=45)
+    ax.set_title(f"Mean: {mean_score:.2f}, Mode: {mode_str}", fontsize=40)
+    ax.set_xlabel("Score", fontsize=34)
+    ax.set_ylabel("Frequency", fontsize=34)
+    ax.tick_params(axis='x', rotation=45, labelsize=34) 
+    ax.tick_params(axis='y', labelsize=34) 
 
 plt.tight_layout()
-plt.savefig("Prometheus Evaluation Results - Base Model - spaced.pdf")
+plt.savefig("Prometheus Evaluation Results - Finetuned Model - spaced.pdf")
 plt.show()

inference/llama_finetuned_inference.py

@@ -29,7 +29,7 @@ for example in reference_dataset:
     code_no += 1
     content = f'''
     Refine the C# code enclosed within tags [C#] and [/C#]. 
-    Provide the refined code enclosed within tags [refined_C#] and [/refined_C#]
+    Provide the refined code must be enclosed within tags [refined_C#] and [/refined_C#]
     The summary of changes must be enclosed within tags [code_changes] and [/code_changes].
     
     [C#]
@@ -65,7 +65,6 @@ for example in reference_dataset:
         output = output.split('<|start_header_id|>assistant<|end_header_id|>')
         if len(output) == 2:
             output = output[1]
-            print(output)
             
             code_pattern = r'\[refined_C#\](.*?)\[/refined_C#\]'
             summary_pattern = r'\[code_changes\](.*?)\[/code_changes\]'