Update app.py

app.py

@@ -77,18 +77,22 @@ zerogpu_fix_js = """
                     
                     const originalFetch = window.fetch;
                     window.fetch = async function(url, options) {
-                        if (!options) options = {};
-                        if (!options.headers) options.headers = {};
-                        
-                        const headerMap = headers instanceof Map ? headers : new Map(Object.entries(headers));
-                        
-                        headerMap.forEach((value, key) => {
-                             if (options.headers instanceof Headers) {
-                                 options.headers.append(key, value);
-                             } else {
-                                 options.headers[key] = value;
-                             }
-                        });
+                        // Only inject headers for prediction requests to avoid rate limiting on heartbeats/static assets
+                        // Gradio requests typically go to /queue/join, /api/predict, etc.
+                        if (typeof url === 'string' && (url.includes('/queue/join') || url.includes('/call/') || url.includes('/api/'))) {
+                             if (!options) options = {};
+                             if (!options.headers) options.headers = {};
+                             
+                             const headerMap = headers instanceof Map ? headers : new Map(Object.entries(headers));
+                             
+                             headerMap.forEach((value, key) => {
+                                  if (options.headers instanceof Headers) {
+                                      options.headers.append(key, value);
+                                  } else {
+                                      options.headers[key] = value;
+                                  }
+                             });
+                        }
                         
                         return originalFetch(url, options);
                     };
@@ -402,7 +406,8 @@ async def segmentation(request: Request):
         return {"error": str(e)}
 
 # Mount the Gradio app
-demo.queue(default_concurrency_limit=20, max_size=40)
+# Reduce concurrency for ZeroGPU to prevent rate limiting
+demo.queue(default_concurrency_limit=10, max_size=20)
 app = gr.mount_gradio_app(app, demo, path="/", root_path="/demo")
 
 if __name__ == "__main__":