Update app.py

app.py

@@ -12,11 +12,8 @@ load_dotenv()
 HF_API_KEY = os.getenv("HF_API_KEY")
 OPENROUTER_API_KEY = os.getenv("OPENROUTER_API_KEY")
 
-# Initialize Hugging Face Gemma Client
-hf_client = InferenceClient(
-    provider="hf-inference",
-    api_key=HF_API_KEY
-)
+# Initialize Hugging Face Clients
+hf_client = InferenceClient(provider="hf-inference", api_key=HF_API_KEY)
 
 # Initialize OpenRouter DeepSeek Client
 openrouter_client = OpenAI(
@@ -24,20 +21,20 @@ openrouter_client = OpenAI(
     api_key=OPENROUTER_API_KEY
 )
 
-# Function to query Gemma-2-27B (Hugging Face)
-def query_gemma(user_input):
+# Query Hugging Face Models
+def query_huggingface_model(user_input, model_name):
     try:
         messages = [{"role": "user", "content": user_input}]
         completion = hf_client.chat.completions.create(
-            model="google/gemma-2-27b-it", 
+            model=model_name, 
             messages=messages, 
             max_tokens=500
         )
         return completion.choices[0].message["content"]
     except Exception as e:
-        return f"Error querying Gemma: {str(e)}"
+        return f"Error querying {model_name}: {str(e)}"
 
-# Function to query DeepSeek-R1 (OpenRouter)
+# Query DeepSeek-R1 (OpenRouter)
 def query_deepseek(user_input):
     try:
         completion = openrouter_client.chat.completions.create(
@@ -51,24 +48,31 @@ def query_deepseek(user_input):
 # Function to refine responses using DeepSeek
 def refine_response(user_input):
     try:
-        # Get responses from both models
-        gemma_response = query_gemma(user_input)
+        # Get responses from all three models
+        gemma_response = query_huggingface_model(user_input, "google/gemma-2-27b-it")
+        llama_response = query_huggingface_model(user_input, "meta-llama/Llama-3.3-70B-Instruct")
         deepseek_response = query_deepseek(user_input)
 
-        # If either response failed, return the available one
-        if not gemma_response.strip():
-            return f"Only DeepSeek Response:\n{deepseek_response}"
-        if not deepseek_response.strip():
-            return f"Only Gemma Response:\n{gemma_response}"
+        # If any response is missing, return the available ones
+        responses = {
+            "Gemma": gemma_response.strip(),
+            "Llama": llama_response.strip(),
+            "DeepSeek": deepseek_response.strip()
+        }
+        valid_responses = {k: v for k, v in responses.items() if v}
+
+        if len(valid_responses) < 2:
+            return "\n\n".join(f"{k} Response: {v}" for k, v in valid_responses.items())
 
         # Prepare refinement prompt
         improvement_prompt = f"""
-        Here are two AI-generated responses:
+        Here are three AI-generated responses:
 
         Response 1 (Gemma): {gemma_response}
-        Response 2 (DeepSeek): {deepseek_response}
+        Response 2 (Llama 3.3): {llama_response}
+        Response 3 (DeepSeek): {deepseek_response}
 
-        Please combine the best elements of both, improve clarity, and provide a final refined answer.
+        Please combine the best elements of all three, improve clarity, and provide a final refined answer.
         """
 
         # Retry loop for OpenRouter API
@@ -85,31 +89,24 @@ def refine_response(user_input):
                         "model": "deepseek/deepseek-r1:free",
                         "messages": [{"role": "user", "content": improvement_prompt}]
                     },
-                    timeout=30  # Increase timeout to 30 seconds
+                    timeout=30
                 )
 
-                # Debugging: Print response
                 print(f"Attempt {attempt + 1}: OpenRouter Response:", response.text)
-
-                # Check if the response is valid JSON
                 response_json = response.json()
-                
-                # Extract the refined response
                 refined_content = response_json["choices"][0]["message"]["content"]
 
-                # If DeepSeek gives an empty response, retry
                 if refined_content.strip():
                     return refined_content
                 else:
                     print("Received empty response from DeepSeek, retrying...")
-                    time.sleep(2)  # Wait before retrying
+                    time.sleep(2)
 
             except Exception as e:
                 print(f"Error on attempt {attempt + 1}: {str(e)}")
-                time.sleep(2)  # Wait before retrying
+                time.sleep(2)
 
-        # If all retries fail, return the better of the two initial responses
-        return f"Refinement failed. Here’s the best available response:\n\n{max(gemma_response, deepseek_response, key=len)}"
+        return f"Refinement failed. Here’s the best available response:\n\n{max(valid_responses.values(), key=len)}"
 
     except Exception as e:
         return f"Error refining response: {str(e)}"
@@ -119,8 +116,8 @@ iface = gr.Interface(
     fn=refine_response, 
     inputs=gr.Textbox(lines=2, placeholder="Ask me anything..."), 
     outputs="text",
-    title="AI Response Enhancer",
-    description="Get responses from both Gemma and DeepSeek, then receive an improved final answer."
+    title="Multi-Model AI Enhancer",
+    description="Get responses from Gemma, Llama 3.3, and DeepSeek. Then receive an improved final answer."
 )
 
 # Launch app