v2.0: Proper code extraction for base model

scripts/eval_bigcodebench_hf.py

@@ -5,6 +5,8 @@
 """
 BigCodeBench Evaluation: Base Devstral vs Fine-tuned Alizee-Coder
 Runs on HF Jobs with GPU support
+
+VERSION: 2.0 - Proper code extraction for both base and fine-tuned models
 """
 
 import os
@@ -102,7 +104,7 @@ def extract_python_code(text):
     return text.strip()
 
 def generate_completion_base(model, tokenizer, prompt):
-    """Generate code completion for BASE model (direct completion)"""
+    """Generate code completion for BASE model (handles chat-like responses)"""
     inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=4096).to(model.device)
 
     with torch.no_grad():
@@ -115,7 +117,27 @@ def generate_completion_base(model, tokenizer, prompt):
             eos_token_id=tokenizer.eos_token_id,
         )
 
-    completion = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
+    raw_completion = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], skip_special_tokens=True)
+
+    # Try to extract code from ```python blocks (if model generates chat-like response)
+    completion = extract_python_code(raw_completion)
+
+    # If extracted code looks like a full function, extract just the body
+    if completion.strip().startswith("def "):
+        lines = completion.split('\n')
+        body_lines = []
+        in_function = False
+        for line in lines:
+            if line.strip().startswith("def "):
+                in_function = True
+                continue
+            if in_function:
+                body_lines.append(line)
+        if body_lines:
+            completion = '\n'.join(body_lines)
+    # If no code block, use raw completion
+    elif completion == raw_completion.strip():
+        completion = raw_completion
 
     # Stop at function boundary
     stop_tokens = ["\ndef ", "\nclass ", "\nif __name__", "\n\n\n"]