Auto-update via watch script

app.py

@@ -162,40 +162,44 @@ def respond(message, history, system_prompt, max_tokens, temperature, disable_th
         prompt_parts.append(f"<|user|>\n{img_tags}{user_text}<|end|>\n<|assistant|>\n")
         prompt = "".join(prompt_parts)
         
-        # Prepare inputs
-        if image_paths:
-            print(f"[inference] Processing prompt with {len(image_paths)} images...", flush=True)
-            images = og.Images.open(*image_paths)
-            inputs = processor(prompt, images=images)
-        else:
-            print("[inference] Processing text-only prompt for Phi-3...", flush=True)
-            try:
-                inputs = processor(prompt)
-            except Exception:
-                inputs = tokenizer.encode(prompt)
+        try:
+            # Prepare inputs
+            if image_paths:
+                print(f"[inference] Processing prompt with {len(image_paths)} images...", flush=True)
+                images = og.Images.open(*image_paths)
+                inputs = processor(prompt, images=images)
+            else:
+                print("[inference] Processing text-only prompt for Phi-3...", flush=True)
+                try:
+                    inputs = processor(prompt)
+                except Exception:
+                    inputs = tokenizer.encode(prompt)
+                    
+            # Set up generator parameters
+            params = og.GeneratorParams(model)
+            if hasattr(inputs, 'set_inputs') or not isinstance(inputs, (list, tuple)):
+                params.set_inputs(inputs)
+            else:
+                params.input_ids = inputs
                 
-        # Set up generator parameters
-        params = og.GeneratorParams(model)
-        if hasattr(inputs, 'set_inputs') or not isinstance(inputs, (list, tuple)):
-            params.set_inputs(inputs)
-        else:
-            params.input_ids = inputs
+            params.set_search_options(
+                max_length=int(max_tokens),
+                temperature=temperature
+            )
             
-        params.set_search_options(
-            max_length=int(max_tokens),
-            temperature=temperature
-        )
-        
-        # Stream output
-        generator = og.Generator(model, params)
-        partial = ""
-        while not generator.is_done():
-            generator.compute_logits()
-            generator.generate_next_token()
-            new_token = generator.get_next_tokens()[0]
-            decoded = tokenizer.decode([new_token])
-            partial += decoded
-            yield partial
+            # Stream output
+            generator = og.Generator(model, params)
+            partial = ""
+            while not generator.is_done():
+                generator.compute_logits()
+                generator.generate_next_token()
+                new_token = generator.get_next_tokens()[0]
+                decoded = tokenizer.decode([new_token])
+                partial += decoded
+                yield partial
+        except Exception as e:
+            print(f"[inference] Error during Phi-3 generation: {e}", flush=True)
+            yield f"⚠️ **Error during model inference**: {str(e)}\n\n*This was caught gracefully to prevent the Space from crashing. Please try again or rephrase.*"
     else:
         # Qwen modes
         llm = _load_llm(mode)