Update app.py

app.py

@@ -110,55 +110,100 @@ def conversation(qa_chain, message, history):
     return qa_chain, gr.update(value=""), new_history, response_source1, response_source1_page, response_source2, response_source2_page
 
 def upload_file(file_obj):
-    list_file_path = [file.name for file in file_obj]
+    list_file_path = []
+    for idx, file in enumerate(file_obj):
+        file_path = file_obj.name
+        list_file_path.append(file_path)
     return list_file_path
 
 
-def gradio_ui():
+def demo():
     with gr.Blocks(theme="base") as demo:
-        # States
-        vector_db, qa_chain, collection_name = gr.State(), gr.State(), gr.State()
-        db_progress, llm_progress = gr.Textbox(), gr.Textbox()
-        chatbot, doc_source1, source1_page, doc_source2, source2_page = gr.Chatbot(), gr.Textbox(), gr.Number(), gr.Textbox(), gr.Number()
-        msg = gr.Textbox(placeholder="Type message")
-
-        with gr.Tabs():
-            # Tab 1: Document Pre-processing
-            with gr.Tab("Step 1 - Document Pre-processing"):
+        vector_db = gr.State()
+        qa_chain = gr.State()
+        collection_name = gr.State()
+        
+        gr.Markdown(
+        """<center><h2>PDF-based chatbot (powered by LangChain and open-source LLMs)</center></h2>
+        <h3>Ask any questions about your PDF documents, along with follow-ups</h3>
+        <b>Note:</b> This AI assistant performs retrieval-augmented generation from your PDF documents. \
+        When generating answers, it takes past questions into account (via conversational memory), and includes document references for clarity purposes.</i>
+        <br><b>Warning:</b> This space uses the free CPU Basic hardware from Hugging Face. Some steps and LLM models used below (free inference endpoints) can take some time to generate an output.<br>
+        """)
+        with gr.Tab("Step 1 - Document pre-processing"):
+            with gr.Row():
+                document = gr.Files(height=100, file_count="multiple", file_types=["pdf"], interactive=True, label="Upload your PDF documents (single or multiple)")
+                # upload_btn = gr.UploadButton("Loading document...", height=100, file_count="multiple", file_types=["pdf"], scale=1)
+            with gr.Row():
+                db_btn = gr.Radio(["ChromaDB"], label="Vector database type", value = "ChromaDB", type="index", info="Choose your vector database")
+            with gr.Accordion("Advanced options - Document text splitter", open=False):
                 with gr.Row():
-                    document = gr.File(label="Upload your PDF document", file_types=["pdf"])
+                    slider_chunk_size = gr.Slider(minimum = 100, maximum = 1000, value=600, step=20, label="Chunk size", info="Chunk size", interactive=True)
                 with gr.Row():
-                    chunk_size = gr.Slider(minimum=100, maximum=1000, value=600, step=50, label="Chunk size", interactive=True)
-                    chunk_overlap = gr.Slider(minimum=10, maximum=200, value=50, step=10, label="Chunk overlap", interactive=True)
+                    slider_chunk_overlap = gr.Slider(minimum = 10, maximum = 200, value=40, step=10, label="Chunk overlap", info="Chunk overlap", interactive=True)
+            with gr.Row():
+                db_progress = gr.Textbox(label="Vector database initialization", value="None")
+            with gr.Row():
+                db_btn = gr.Button("Generate vector database...")
+            
+        with gr.Tab("Step 2 - QA chain initialization"):
+            with gr.Row():
+                llm_btn = gr.Radio(list_llm_simple, \
+                    label="LLM models", value = list_llm_simple[0], type="index", info="Choose your LLM model")
+            with gr.Accordion("Advanced options - LLM model", open=False):
                 with gr.Row():
-                    db_init_btn = gr.Button("Initialize Vector Database")
-
-            # Tab 2: QA Chain Initialization
-            with gr.Tab("Step 2 - QA Chain Initialization"):
-                with gr.Row():
-                    llm_selection = gr.Radio(list_llm_simple, label="Choose LLM Model", value=list_llm_simple[0])
+                    slider_temperature = gr.Slider(minimum = 0.0, maximum = 1.0, value=0.7, step=0.1, label="Temperature", info="Model temperature", interactive=True)
                 with gr.Row():
-                    temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.7, step=0.1, label="Temperature", interactive=True)
-                    max_tokens = gr.Slider(minimum=64, maximum=1024, value=256, step=64, label="Max Tokens", interactive=True)
-                    top_k = gr.Slider(minimum=1, maximum=10, value=3, step=1, label="Top K", interactive=True)
+                    slider_maxtokens = gr.Slider(minimum = 224, maximum = 4096, value=1024, step=32, label="Max Tokens", info="Model max tokens", interactive=True)
                 with gr.Row():
-                    qa_init_btn = gr.Button("Initialize QA Chain")
-
-            # Tab 3: Conversation with Chatbot
-            with gr.Tab("Step 3 - Conversation with Chatbot"):
-                chat_history = gr.State()
+                    slider_topk = gr.Slider(minimum = 1, maximum = 10, value=3, step=1, label="top-k samples", info="Model top-k samples", interactive=True)
+            with gr.Row():
+                llm_progress = gr.Textbox(value="None",label="QA chain initialization")
+            with gr.Row():
+                qachain_btn = gr.Button("Initialize question-answering chain...")
+
+        with gr.Tab("Step 3 - Conversation with chatbot"):
+            chatbot = gr.Chatbot(height=300)
+            with gr.Accordion("Advanced - Document references", open=False):
                 with gr.Row():
-                    chatbot
+                    doc_source1 = gr.Textbox(label="Reference 1", lines=2, container=True, scale=20)
+                    source1_page = gr.Number(label="Page", scale=1)
                 with gr.Row():
-                    msg
-                    submit_btn = gr.Button("Submit")
-
-        # Handlers
-        db_init_btn.click(initialize_database, inputs=[document, chunk_size, chunk_overlap], outputs=[vector_db, collection_name, db_progress])
-        qa_init_btn.click(initialize_LLM, inputs=[llm_selection, temperature, max_tokens, top_k, vector_db], outputs=[qa_chain, llm_progress])
-        submit_btn.click(conversation, inputs=[qa_chain, msg, chat_history], outputs=[qa_chain, msg, chatbot, doc_source1, source1_page, doc_source2, source2_page])
+                    doc_source2 = gr.Textbox(label="Reference 2", lines=2, container=True, scale=20)
+                    source2_page = gr.Number(label="Page", scale=1)
+            with gr.Row():
+                msg = gr.Textbox(placeholder="Type message", container=True)
+            with gr.Row():
+                submit_btn = gr.Button("Submit")
+                clear_btn = gr.ClearButton([msg, chatbot])
+            
+        # Preprocessing events
+        #upload_btn.upload(upload_file, inputs=[upload_btn], outputs=[document])
+        db_btn.click(initialize_database, \
+            inputs=[document, slider_chunk_size, slider_chunk_overlap], \
+            outputs=[vector_db, collection_name, db_progress])
+        qachain_btn.click(initialize_LLM, \
+            inputs=[llm_btn, slider_temperature, slider_maxtokens, slider_topk, vector_db], \
+            outputs=[qa_chain, llm_progress]).then(lambda:[None,"",0,"",0], \
+            inputs=None, \
+            outputs=[chatbot, doc_source1, source1_page, doc_source2, source2_page], \
+            queue=False)
+
+        # Chatbot events
+        msg.submit(conversation, \
+            inputs=[qa_chain, msg, chatbot], \
+            outputs=[qa_chain, msg, chatbot, doc_source1, source1_page, doc_source2, source2_page], \
+            queue=False)
+        submit_btn.click(conversation, \
+            inputs=[qa_chain, msg, chatbot], \
+            outputs=[qa_chain, msg, chatbot, doc_source1, source1_page, doc_source2, source2_page], \
+            queue=False)
+        clear_btn.click(lambda:[None,"",0,"",0], \
+            inputs=None, \
+            outputs=[chatbot, doc_source1, source1_page, doc_source2, source2_page], \
+            queue=False)
+    demo.queue().launch(debug=True)
 
-    return demo
 
 if __name__ == "__main__":
-    gradio_ui().launch()
+    demo()