fix: update device mapping and GPU duration, and add thinking UI state to chat interface

app.py

@@ -17,7 +17,7 @@ model = AutoModelForImageTextToText.from_pretrained(
     model_id, 
     torch_dtype=torch.bfloat16, 
     trust_remote_code=True,
-    device_map="auto"
+    device_map="cuda"
 )
 
 # Utility for response normalization
@@ -33,8 +33,8 @@ def normalize_response_text(text: str) -> str:
 
 app = Server()
 
-@spaces.GPU
 @app.api()
+@spaces.GPU(duration=120)
 def predict(message: str, file: FileData = None, downsample_mode: str = "16x"):
     """
     General inference endpoint for both image and video.

index.html

@@ -294,27 +294,49 @@
             
             addMessage('user', currentText, fileUrl, fileType);
             
-            // Show loading
+            // Show loading state
             sendIcon.classList.add('hidden');
             loadingSpinner.classList.remove('hidden');
             sendBtn.disabled = true;
 
+            // Add thinking placeholder
+            const thinkingDiv = document.createElement('div');
+            thinkingDiv.className = 'flex gap-4 max-w-3xl mx-auto items-start message-anim';
+            thinkingDiv.id = 'thinking-placeholder';
+            thinkingDiv.innerHTML = `
+                <div class="w-8 h-8 rounded-full bg-indigo-500/20 flex items-center justify-center shrink-0 border border-indigo-500/30">
+                    <i data-lucide="bot" class="w-4 h-4 text-indigo-400"></i>
+                </div>
+                <div class="glass p-5 rounded-2xl rounded-tl-none border border-white/5">
+                    <div class="flex items-center gap-2 text-gray-400 italic">
+                        <i data-lucide="loader-2" class="w-3 h-3 animate-spin"></i> MiniCPM is thinking...
+                    </div>
+                </div>
+            `;
+            chatMessages.appendChild(thinkingDiv);
+            lucide.createIcons();
+            chatMessages.scrollTop = chatMessages.scrollHeight;
+
             try {
                 let fileData = null;
                 if (currentFile) {
                     fileData = handle_file(currentFile);
                 }
 
+                console.log("Sending request to MiniCPM-V 4.6 backend...");
                 const result = await client.predict("/predict", {
                     message: currentText,
                     file: fileData,
                     downsample_mode: currentMode
                 });
 
+                // Remove placeholder and add actual message
+                document.getElementById('thinking-placeholder')?.remove();
                 addMessage('assistant', result.data);
             } catch (error) {
                 console.error("Prediction failed:", error);
-                addMessage('assistant', "Sorry, I encountered an error while processing your request.");
+                document.getElementById('thinking-placeholder')?.remove();
+                addMessage('assistant', "Sorry, I encountered an error while processing your request. Please check the Space logs.");
             } finally {
                 sendIcon.classList.remove('hidden');
                 loadingSpinner.classList.add('hidden');