feat: Add GROQ_TOPIC_MODEL configuration and enhance TopicService to utilize it for improved model selection during initialization, including fallback options for robustness.

config.py

@@ -42,6 +42,9 @@ GROQ_TTS_FORMAT = "wav"
 # **Chat Model**
 GROQ_CHAT_MODEL = "llama3-70b-8192"
 
+# **Topic Extraction Model**
+GROQ_TOPIC_MODEL = "llama-3.1-70b-versatile"  # Alternative: "llama3-70b-8192" or "llama-3.1-8b-instant"
+
 # ============ SUPABASE ============
 SUPABASE_URL = os.getenv("SUPABASE_URL", "")
 SUPABASE_KEY = os.getenv("SUPABASE_KEY", "")

services/topic_service.py

@@ -7,7 +7,7 @@ from langchain_groq import ChatGroq
 from pydantic import BaseModel, Field
 from langsmith import traceable
 
-from config import GROQ_API_KEY
+from config import GROQ_API_KEY, GROQ_TOPIC_MODEL
 
 logger = logging.getLogger(__name__)
 
@@ -22,7 +22,15 @@ class TopicService:
     
     def __init__(self):
         self.llm = None
-        self.model_name = "openai/gpt-oss-safeguard-120b"  # another model meta-llama/llama-4-scout-17b-16e-instruct
+        # Use valid Groq model - defaults from config, fallback to common models
+        self.model_name = GROQ_TOPIC_MODEL if GROQ_TOPIC_MODEL else "llama-3.1-70b-versatile"
+        # Fallback models to try if primary fails
+        self.fallback_models = [
+            "llama-3.1-70b-versatile",
+            "llama3-70b-8192",
+            "llama-3.1-8b-instant",
+            "mixtral-8x7b-32768"
+        ]
         self.initialized = False
         
     def initialize(self, model_name: Optional[str] = None):
@@ -37,25 +45,37 @@ class TopicService:
         if model_name:
             self.model_name = model_name
             
-        try:
-            logger.info(f"Initializing topic extraction service with model: {self.model_name}")
-            
-            llm = ChatGroq(
-                model=self.model_name,
-                api_key=GROQ_API_KEY,
-                temperature=0.0,
-                max_tokens=512,
-            )
-            
-            # Bind structured output directly to the model
-            self.llm = llm.with_structured_output(TopicOutput)
-            self.initialized = True
-            
-            logger.info("✓ Topic extraction service initialized successfully")
-            
-        except Exception as e:
-            logger.error(f"Error initializing topic service: {str(e)}")
-            raise RuntimeError(f"Failed to initialize topic service: {str(e)}")
+        # Try primary model first, then fallbacks
+        models_to_try = [self.model_name] + [m for m in self.fallback_models if m != self.model_name]
+        
+        last_error = None
+        for model_to_try in models_to_try:
+            try:
+                logger.info(f"Initializing topic extraction service with model: {model_to_try}")
+                
+                llm = ChatGroq(
+                    model=model_to_try,
+                    api_key=GROQ_API_KEY,
+                    temperature=0.0,
+                    max_tokens=512,
+                )
+                
+                # Bind structured output directly to the model
+                self.llm = llm.with_structured_output(TopicOutput)
+                self.model_name = model_to_try  # Update to successful model
+                self.initialized = True
+                
+                logger.info(f"✓ Topic extraction service initialized successfully with model: {model_to_try}")
+                return
+                
+            except Exception as e:
+                last_error = e
+                logger.warning(f"Failed to initialize with model {model_to_try}: {str(e)}")
+                continue
+        
+        # If all models failed
+        logger.error(f"Error initializing topic service with all models: {last_error}")
+        raise RuntimeError(f"Failed to initialize topic service with any model. Last error: {str(last_error)}")
     
     @traceable(name="extract_topic")
     def extract_topic(self, text: str) -> str: