update model to gemma-3-270m-it

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-2b-it via HuggingFace Inference API.
+    Generates a tailored CVE summary using google/gemma-3-270m-it via HuggingFace Inference API.
     
     Args:
         cve_description: The original CVE description
@@ -382,7 +382,7 @@ def generate_tailored_summary(cve_description: str, audience: str, hf_token: Opt
     
     # Define the model to use
     models = [
-        "google/gemma-2b-it",
+        "google/gemma-3-270m-it",
     ]
     
     headers = {"Authorization": f"Bearer {token}"}
@@ -588,7 +588,7 @@ 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 SmolLM3-3B model.
+                            - AI-powered audience-specific summaries using the google/gemma-3-270m-it model.
                             
                             **Supported Audiences:**
                             - **Cybersecurity Professional:** Focus on threats, attack vectors, and mitigation
@@ -600,9 +600,11 @@ def create_interface():
                             
                             **Data Source:** [NIST NVD API](https://nvd.nist.gov/developers/vulnerabilities)
                             
-                            **AI Model:** [google/gemma-2b-it](https://huggingface.co/google/gemma-2b-it)
+                            **AI Model:** [google/gemma-3-270m-it](https://huggingface.co/google/gemma-3-270m-it)
 
                             **Disclaimer:** Generated content may be inaccurate or false.
+
+                            The free community tier of the Hugging Face Inference API powers this app's AI features. Since computing resources are shared, anticipate some delay on your initial request as the model loads. Later requests usually process more quickly.
                             
                             **Performance Optimizations:**
                             - Shorter timeouts for faster failure detection