Update tinyllama_inference.py with improved evaluation and performance

tinyllama_inference.py

@@ -8,13 +8,14 @@ tokenizer, model = None, None
 def load_model():
     global tokenizer, model
     if tokenizer is None or model is None:
-        model_name = "Salesforce/codegen-350M-mono"  # Public model for code evaluation
+        # Use a DeepSeek model for code evaluation.
+        model_name = "deepseek-ai/deepseek-coder-1.3b"  # Adjust to your chosen DeepSeek model
         tokenizer = AutoTokenizer.from_pretrained(model_name)
         model = AutoModelForCausalLM.from_pretrained(model_name)
     return tokenizer, model
 
 def evaluate_code(question, code):
-    # Updated prompt: explicitly instructs the model to output exactly valid JSON
+    # Updated prompt: instructs the model to output exactly valid JSON
     prompt = f"""You are an expert code evaluator.
 Evaluate the following solution for the given problem.
 Respond with exactly one JSON object (with no extra text) that has exactly two keys:
@@ -27,20 +28,20 @@ Solution: "{code}"
 """
     tokenizer, model = load_model()
     inputs = tokenizer(prompt, return_tensors="pt")
-    # Adjust parameters for more deterministic and concise output
+    # Adjust parameters for concise and deterministic output
     outputs = model.generate(
         **inputs,
-        max_new_tokens=60,     # Fewer tokens to limit output length
+        max_new_tokens=60,     # Limit output length
         temperature=0.0,       # Deterministic output
         pad_token_id=tokenizer.eos_token_id,
         do_sample=False
     )
     response_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
     
-    # Debug: print the raw response to logs if needed
+    # Debug: Uncomment the following line to print raw output if needed
     # print("Raw model response:", response_text)
     
-    # Use regex to extract the first JSON object that starts with '{' and ends with '}'
+    # Use regex to extract the JSON object from the response
     match = re.search(r'\{.*\}', response_text)
     if match:
         json_text = match.group(0)