Update app.py

app.py

@@ -47,18 +47,18 @@ def text_to_image_generate(input_prompt, output_path, vl_chat_processor, vl_gpt,
 
     inputs_embeds = vl_gpt.language_model.get_input_embeddings()(tokens)
     generated_tokens = torch.zeros((parallel_size, image_token_num), dtype=torch.int).cuda()
-    past_key_values = None
 
     for i in range(image_token_num):
         if i == 0:
-            outputs = vl_gpt.language_model.model(inputs_embeds=inputs_embeds, use_cache=True)
+            current_inputs = inputs_embeds
         else:
-            outputs = vl_gpt.language_model.model(
-                inputs_embeds=img_embeds.unsqueeze(1),
-                use_cache=True,
-                past_key_values=past_key_values
-            )
-        past_key_values = outputs.past_key_values
+            current_inputs = img_embeds.unsqueeze(1)
+
+        # ✅ No past_key_values, crash-safe
+        outputs = vl_gpt.language_model.model(
+            inputs_embeds=current_inputs,
+            use_cache=False
+        )
 
         hidden_states = outputs.last_hidden_state
         logits = vl_gpt.gen_head(hidden_states[:, -1, :])
@@ -117,4 +117,4 @@ demo = gr.ChatInterface(
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()