feat: implement dynamic stream timeout to improve response handling responsiveness

index.html

@@ -536,15 +536,17 @@
 
                 const iterator = submission[Symbol.asyncIterator]();
                 let streamDone = false;
+                let hasReceivedData = false;
 
                 while (!streamDone) {
-                    // Race each iterator.next() against a 30s idle timeout.
-                    // The Gradio client's iterator may hang forever after the
-                    // last event, so this guarantees we always break out.
+                    // Dynamic timeout: 60s before first data (queue/connection wait),
+                    // 2s after first data (idle = stream finished, since tokens
+                    // arrive every few hundred ms during generation).
+                    const idleTimeout = hasReceivedData ? 2000 : 60000;
                     const result = await Promise.race([
                         iterator.next(),
                         new Promise(resolve =>
-                            setTimeout(() => resolve({ value: undefined, done: true }), 30000)
+                            setTimeout(() => resolve({ value: undefined, done: true }), idleTimeout)
                         )
                     ]);
 
@@ -556,6 +558,7 @@
                     const event = result.value;
 
                     if (event.type === "data") {
+                        hasReceivedData = true;
                         typingIndicatorWrapper.style.display = 'none';
                         typingIndicator.style.display = 'none';
                         fullResponse = event.data[0];