Update app/core/model_loader.py

app/core/model_loader.py

@@ -13,6 +13,11 @@ load_dotenv()
 logger = logging.getLogger(__name__)
 STATS_FILE = "usage_stats.json"
 
+# Supported active Groq free-tier models (Fallback list)
+DEFAULT_MODEL = "llama-3.1-8b-instant"
+FALLBACK_MODEL = "llama-3.3-70b-versatile"
+
+
 class LLMSingleton:
     _instance = None
     _instance_lock = threading.Lock()
@@ -29,21 +34,26 @@ class LLMSingleton:
         if self._instance is not None:
             raise Exception("Singleton instance already exists!")
 
-        self.api_key = os.getenv("GROQ_API_KEY")
+        # 1. Load and sanitize API Key
+        raw_key = os.getenv("GROQ_API_KEY", "")
+        self.api_key = raw_key.strip().strip('"').strip("'") if raw_key else ""
+
         if not self.api_key:
-            logger.warning("⚠️ GROQ_API_KEY not found in environment variables. AI features will fail.")
+            logger.warning("⚠️ GROQ_API_KEY not found or empty. AI features will fail.")
             self.client = None
         else:
             self.client = Groq(api_key=self.api_key)
 
-        # Sanitize model string to strip hidden whitespace, newlines, or quotes injected by HF UI
-        raw_model = os.getenv("GROQ_MODEL", "llama-3.1-8b-instant")
+        # 2. Load and sanitize Model Name
+        raw_model = os.getenv("GROQ_MODEL", DEFAULT_MODEL)
         clean_model = raw_model.strip().strip('"').strip("'") if raw_model else ""
-        self.model_name = clean_model if clean_model else "llama-3.1-8b-instant"
+        self.model_name = clean_model if clean_model else DEFAULT_MODEL
 
+        logger.info(f"🔑 Groq Client initialized with model target: {self.model_name}")
+
+        # 3. Threading locks and usage stats setup
         self._stats_lock = threading.Lock()
         self._rpm_lock = threading.Lock()
-
         self.stats = self._load_stats()
         self._check_daily_reset()
 
@@ -125,6 +135,7 @@ class LLMSingleton:
             self.stats["total_requests"] += 1
             self.stats["daily_requests_count"] += 1
             self._save_stats()
+
         with self._rpm_lock:
             self.requests_this_minute += 1
         return True
@@ -146,6 +157,7 @@ class LLMSingleton:
         retries = 0
         max_retries = 3
         base_delay = 2
+        active_model = self.model_name
 
         while retries <= max_retries:
             try:
@@ -154,7 +166,7 @@ class LLMSingleton:
                     self._save_stats()
 
                 chat_completion = self.client.chat.completions.create(
-                    model=self.model_name,
+                    model=active_model,
                     messages=[
                         {
                             "role": "system",
@@ -179,10 +191,22 @@ class LLMSingleton:
 
             except Exception as e:
                 error_str = str(e)
+
+                # Automatic Model Fallback handling if a 404/model_not_found occurs
+                if "404" in error_str or "model_not_found" in error_str.lower():
+                    if active_model != FALLBACK_MODEL:
+                        logger.warning(
+                            f"⚠️ Model target '{active_model}' rejected (404). Falling back to '{FALLBACK_MODEL}'..."
+                        )
+                        active_model = FALLBACK_MODEL
+                        continue
+
+                # Rate Limit handling (HTTP 429)
                 if "429" in error_str or "rate_limit_exceeded" in error_str.lower():
                     with self._stats_lock:
                         self.stats["rate_limit_hits"] += 1
                         self._save_stats()
+
                     wait_time = (base_delay * (2**retries)) + random.uniform(0.5, 1.5)
                     logger.warning(
                         f"⚠️ Groq rate limit hit. Retrying in {wait_time:.2f}s... (Attempt {retries + 1}/{max_retries})"
@@ -199,6 +223,7 @@ class LLMSingleton:
         with self._stats_lock:
             self.stats["errors"] += 1
             self._save_stats()
+
         logger.error("❌ Max retries reached. Request failed.")
         raise RuntimeError("RATE_LIMIT_EXCEEDED")
 
@@ -238,8 +263,11 @@ class LLMSingleton:
                 self._save_stats()
 
             return response_text
+
         except Exception as e:
             logger.error(f"Groq text generation failed: {e}")
             return f"Error generating content: {str(e)}"
 
+
+# Export global instance
 llm_engine = LLMSingleton.get_instance()