v3.0: Proper code extraction for base model (handles chat responses)

scripts/eval_mbpp_hf.py

@@ -6,8 +6,10 @@
 MBPP Evaluation: Base Devstral vs Fine-tuned Alizee-Coder
 Runs on HF Jobs with GPU support
 
-VERSION: 2.0 - Force function renaming before test execution
-FIXED: Extract expected function name and rename generated function to match
+VERSION: 3.0 - Proper code extraction for both base and fine-tuned models
+FIXED:
+- Extract code from ```python blocks for base model (handles chat-like responses)
+- Function renaming before test execution for both models
 """
 
 import os
@@ -124,12 +126,12 @@ def extract_python_code(text):
     return text.strip()
 
 def generate_completion_base(model, tokenizer, prompt, func_name=None):
-    """Generate code completion for BASE model (direct completion)"""
-    # Include expected function name in prompt if available
+    """Generate code completion for BASE model (handles both pure completion and chat-like responses)"""
+    # Use a simple code completion prompt
     if func_name:
-        code_prompt = f"# Python function\n# Task: {prompt}\n# Function name: {func_name}\n\ndef {func_name}("
+        code_prompt = f"# Task: {prompt}\n# Write a Python function named {func_name}\n\n"
     else:
-        code_prompt = f"# Python function\n{prompt}\n\ndef"
+        code_prompt = f"# Task: {prompt}\n\n"
 
     inputs = tokenizer(code_prompt, return_tensors="pt", truncation=True, max_length=2048).to(model.device)
 
@@ -145,19 +147,23 @@ def generate_completion_base(model, tokenizer, prompt, func_name=None):
 
     completion = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
 
-    # Reconstruct the function
-    if func_name:
-        full_code = f"def {func_name}(" + completion
-    else:
-        full_code = "def" + completion
+    # Try to extract code from ```python blocks (if model generates chat-like response)
+    code = extract_python_code(completion)
+
+    # If no code block found, try to find a function definition directly
+    if not code.startswith("def "):
+        # Look for function definition in the raw completion
+        match = re.search(r'(def\s+\w+\s*\([^)]*\).*?)(?=\ndef |\nclass |\n```|\Z)', completion, re.DOTALL)
+        if match:
+            code = match.group(1).strip()
 
     # Stop at function boundary
     stop_tokens = ["\ndef ", "\nclass ", "\nif __name__", "\n\n\n"]
     for stop in stop_tokens:
-        if stop in full_code:
-            full_code = full_code[:full_code.index(stop)]
+        if stop in code:
+            code = code[:code.index(stop)]
 
-    return full_code
+    return code
 
 def generate_completion_finetuned(model, tokenizer, prompt, func_name=None):
     """Generate code completion for FINE-TUNED model (Instruct format)"""