Update tinyllama_inference.py to use deepseek-ai/deepseek-coder-1.3b-instruct

tinyllama_inference.py

@@ -8,8 +8,8 @@ tokenizer, model = None, None
 def load_model():
     global tokenizer, model
     if tokenizer is None or model is None:
-        # Use a DeepSeek model for code evaluation.
-        model_name = "deepseek-ai/deepseek-coder-1.3b"  # Adjust to your chosen DeepSeek model if needed
+        # Use the DeepSeek instruct model for code evaluation.
+        model_name = "deepseek-ai/deepseek-coder-1.3b-instruct"
         tokenizer = AutoTokenizer.from_pretrained(model_name)
         model = AutoModelForCausalLM.from_pretrained(model_name)
     return tokenizer, model
@@ -31,15 +31,15 @@ Solution: "{code}"
     # Adjust parameters for concise and deterministic output
     outputs = model.generate(
         **inputs,
-        max_new_tokens=60,     # Limit output length
+        max_new_tokens=60,     # Limit output length for faster responses
         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)
-    print("Raw model response:", response_text)  # Debug output
+    print("Raw model response:", response_text)  # Debug: Inspect raw output
 
-    # Use regex to extract the JSON object from the response
+    # Use regex (non-greedy) to extract the first JSON object from the response
     match = re.search(r'\{.*?\}', response_text)
     if match:
         json_text = match.group(0)