Update app.py

app.py

@@ -204,7 +204,7 @@ def answer_question(images, question):
     global gen_model, gen_tokenizer
     # here each element of images is a tuple of (image_path, None).
     images_ = [Image.open(image[0]).convert('RGB') for image in images]
-    msgs = [{'role': 'user', 'content': [*images_, question]}]
+    msgs = [{'role': 'user', 'content': [question, *images_]}]
     answer = gen_model.chat(
         image=None,
         msgs=msgs,
@@ -212,26 +212,23 @@ def answer_question(images, question):
     )
     print(answer)
     return answer
-    
 
 
 with gr.Blocks() as app:
     gr.Markdown("# MiniCPMV-RAG-PDFQA: Two Vision Language Models Enable End-to-End RAG")
     
     gr.Markdown("""
-- A Vision Language Model Dense Retriever ([MiniCPM-Visual-Embedding](https://huggingface.co/RhapsodyAI/minicpm-visual-embedding-v0)) **directly reads** your PDFs **without need of OCR**, produce **multimodal dense representations** and build your personal library. 
+- A Vision Language Model Dense Retriever ([minicpm-visual-embedding-v0](https://huggingface.co/RhapsodyAI/minicpm-visual-embedding-v0)) **directly reads** your PDFs **without need of OCR**, produce **multimodal dense representations** and build your personal library. 
 
 - **Ask a question**, it retrieve most relavant pages, then [MiniCPM-V-2.6](https://huggingface.co/spaces/openbmb/MiniCPM-V-2_6) will answer your question based on pages recalled, with strong multi-image understanding capability. 
 
-1. It helps you read a long **visually-intensive** or **text-oriented** PDF document and find the pages that answer your question.
+    - It helps you read a long **visually-intensive** or **text-oriented** PDF document and find the pages that answer your question.
 
-2. It helps you build a personal library and retireve book pages from a large collection of books.
+    - It helps you build a personal library and retireve book pages from a large collection of books.
 
-3. It works like a human: read, store, retrieve, and answer with full vision.
+    - It works like a human: read, store, retrieve, and answer with full vision.
 """)
     
-    gr.Markdown("- We **open-sourced** our visual embedding model at [RhapsodyAI/minicpm-visual-embedding-v0](https://huggingface.co/RhapsodyAI/minicpm-visual-embedding-v0)")
-    
     gr.Markdown("- Currently online demo support PDF document with less than 50 pages.")
     
     with gr.Row():
@@ -251,8 +248,6 @@ with gr.Blocks() as app:
         images_output = gr.Gallery(label="Retrieved Pages")
     
     retrieve_button.click(retrieve_gradio, inputs=[kb_id_input, query_input, topk_input], outputs=images_output)
-    
-    gr.Markdown("By using this demo, you agree to share your use data with us for research purpose, to help improve user experience.")
 
     with gr.Row():
         button = gr.Button("Answer Question with Retrieved Pages")
@@ -267,7 +262,8 @@ with gr.Blocks() as app:
         
     upvote_button.click(upvote, inputs=[kb_id_input, query_input], outputs=None)
     downvote_button.click(downvote, inputs=[kb_id_input, query_input], outputs=None)
-    
+
+    gr.Markdown("By using this demo, you agree to share your use data with us for research purpose, to help improve user experience.")
 
 
 app.launch()