adopted to updated endpoint

app.py

@@ -357,7 +357,7 @@ class CVEDashboard:
 
 def generate_tailored_summary(cve_description: str, audience: str, hf_token: Optional[str] = None, max_retries: int = 2) -> str:
     """
-    Generates a tailored CVE summary using google/gemma-3-270m-it via HuggingFace Inference API.
+    Generates a tailored CVE summary using google/gemma-2-9b-it via HuggingFace Inference API.
     
     Args:
         cve_description: The original CVE description
@@ -380,9 +380,9 @@ def generate_tailored_summary(cve_description: str, audience: str, hf_token: Opt
     if audience not in AUDIENCE_PROFILES:
         return f"❌ Unknown audience: {audience}"
     
-    # Define the model to use
+    # Define the model(s) to use
     models = [
-        "google/gemma-3-270m-it",
+        "google/gemma-2-9b-it",
     ]
     
     headers = {"Authorization": f"Bearer {token}"}
@@ -401,22 +401,22 @@ Rewrite this CVE description for a {audience}.
 {cve_description[:1200]}
 Provide a concise, actionable summary (2-3 sentences) highlighting what matters most to this audience. Focus on practical implications and next steps."""
 
-    # Manually apply the Gemma chat template
-    prompt = f"<start_of_turn>user\n{full_prompt}<end_of_turn>\n<start_of_turn>model\n"
+    # Use the OpenAI-compatible messages format
+    messages = [
+        {"role": "user", "content": full_prompt}
+    ]
+    
+    # Use the new, standardized router endpoint
+    api_url = "https://router.huggingface.co/v1/chat/completions"
     
     for model in models:
-        api_url = f"https://api-inference.huggingface.co/models/{model}"
-        
         payload = {
-            "inputs": prompt,
-            "parameters": {
-                "max_new_tokens": 200,
-                "temperature": 0.7,
-                "top_p": 0.95,           # Recommended
-                "do_sample": True,
-                "return_full_text": False,
-                "stop": ["<end_of_turn>", "<start_of_turn>"] # Stop sequences for Gemma
-            }
+            "model": model,
+            "messages": messages,
+            "max_tokens": 250,
+            "temperature": 0.7,
+            "top_p": 0.95,
+            "stop": ["<end_of_turn>", "<start_of_turn>"] # Stop sequences for Gemma
         }
         
         for attempt in range(max_retries):
@@ -424,19 +424,25 @@ Provide a concise, actionable summary (2-3 sentences) highlighting what matters
                 logger.info(f"Generating summary with {model} (attempt {attempt + 1})")
                 
                 response = requests.post(api_url, headers=headers, json=payload, timeout=45)
+                
                 if response.status_code == 200:
                     try:
                         result = response.json()
                         
-                        # Standard response format for this payload type
+                        # New OpenAI-compatible response parsing
                         summary = ""
-                        if isinstance(result, list) and len(result) > 0:
-                            summary = result[0].get('generated_text', '').strip()
+                        if "choices" in result and len(result["choices"]) > 0:
+                            message = result["choices"][0].get("message", {})
+                            summary = message.get("content", "").strip()
                         
                         if summary and len(summary) > 20:
                             logger.info(f"Successfully generated summary with {model}")
                             return f"**{audience} Summary (via {model.split('/')[-1]}):**\n\n{summary}"
-                            
+                        else:
+                            # Handle cases where the model returns an empty summary
+                            logger.warning(f"Model {model} returned an empty or short summary.")
+                            continue # Retry if possible
+
                     except json.JSONDecodeError as e:
                         logger.warning(f"JSON decode error with {model}: {e}")
                         continue
@@ -451,22 +457,22 @@ Provide a concise, actionable summary (2-3 sentences) highlighting what matters
                         continue
                     else:
                         break
-                
                 else:
-                    logger.warning(f"HTTP {response.status_code} with {model}")
+                    error_message = response.json().get("error", response.text)
+                    logger.warning(f"HTTP {response.status_code} with {model}: {error_message}")
+                    # If the model is not found or there's a validation error, don't retry.
+                    if response.status_code in [404, 422]:
+                        return f"❌ Model '{model}' not found or request is invalid. Please check the model name."
                     break
                         
             except requests.exceptions.Timeout:
-                logger.warning(f"Timeout with {model}, trying next model...")
-                break
+                logger.warning(f"Timeout with {model} on attempt {attempt + 1}")
+                if attempt >= max_retries - 1:
+                    break # Break outer loop if all retries failed
             
             except requests.exceptions.RequestException as e:
                 logger.error(f"Request failed with {model}: {e}")
                 break
-            
-            except Exception as e:
-                logger.error(f"Unexpected error with {model}: {e}")
-                break
     
     return "⏳ AI models are currently busy. This can happen during peak usage. Please try again in a few minutes."
 
@@ -588,7 +594,7 @@ def create_interface():
                             - Search CVEs by date range and keywords
                             - Filter by severity levels
                             - Visualize CVE distributions and trends
-                            - AI-powered audience-specific summaries using the google/gemma-3-270m-it model.
+                            - AI-powered audience-specific summaries using the google/gemma-2-9b-it model.
                             
                             **Supported Audiences:**
                             - **Cybersecurity Professional:** Focus on threats, attack vectors, and mitigation
@@ -600,7 +606,7 @@ def create_interface():
                             
                             **Data Source:** [NIST NVD API](https://nvd.nist.gov/developers/vulnerabilities)
                             
-                            **AI Model:** [google/gemma-3-270m-it](https://huggingface.co/google/gemma-3-270m-it)
+                            **AI Model:** [google/gemma-2-9b-it](https://huggingface.co/google/gemma-2-9b-it)
 
                             **Disclaimer:** Generated content may be inaccurate or false.