Update app.py

app.py

@@ -20,15 +20,6 @@ import torch
 import tqdm 
 import accelerate
 
-#Set parameters
-
-llm_model = 'mistralai/Mixtral-8x7B-Instruct-v0.1'
-list_file_obj = '/home/user/app/pdfs/'
-slider_chunk_size = 1024
-slider_chunk_overlap = 128
-temperature = 0.1
-max_tokens = 6000
-top_k = 3
 
 
 # default_persist_directory = './chroma_HF/'
@@ -247,29 +238,64 @@ def demo():
         vector_db = gr.State()
         qa_chain = gr.State()
         collection_name = gr.State()
-
-
-        document = gr.Files(value = '/home/user/app/pdfs/Annual-Report-2022-2023-English_1.pdf',height=100, 
-                            file_count="multiple", file_types=["pdf"], interactive=True, visible=False,
-                            label="Upload your PDF documents (single or multiple)")
-        chatbot = gr.Chatbot(height=300)
-        with gr.Accordion("Advanced - Document references", open=False):
+        
+        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(value = '/home/user/app/pdfs/Annual-Report-2022-2023-English_1.pdf',visible=True,
+                                    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():
+                    slider_chunk_size = gr.Slider(minimum = 100, maximum = 1000, value=600, step=20, label="Chunk size", info="Chunk size", interactive=True)
+                with gr.Row():
+                    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():
-                doc_source1 = gr.Textbox(label="Reference 1", lines=2, container=True, scale=20)
-                source1_page = gr.Number(label="Page", scale=1)
+                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():
+                    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():
+                    slider_maxtokens = gr.Slider(minimum = 224, maximum = 4096, value=1024, step=32, label="Max Tokens", info="Model max tokens", interactive=True)
+                with gr.Row():
+                    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():
+                    doc_source1 = gr.Textbox(label="Reference 1", lines=2, container=True, scale=20)
+                    source1_page = gr.Number(label="Page", scale=1)
+                with gr.Row():
+                    doc_source2 = gr.Textbox(label="Reference 2", lines=2, container=True, scale=20)
+                    source2_page = gr.Number(label="Page", scale=1)
+                with gr.Row():
+                    doc_source3 = gr.Textbox(label="Reference 3", lines=2, container=True, scale=20)
+                    source3_page = gr.Number(label="Page", scale=1)
             with gr.Row():
-                doc_source2 = gr.Textbox(label="Reference 2", lines=2, container=True, scale=20)
-                source2_page = gr.Number(label="Page", scale=1)
+                msg = gr.Textbox(placeholder="Type message", container=True)
             with gr.Row():
-                doc_source3 = gr.Textbox(label="Reference 3", lines=2, container=True, scale=20)
-                source3_page = gr.Number(label="Page", scale=1)
-        with gr.Row():
-            msg = gr.Textbox(placeholder="Type message", container=True)
-        with gr.Row():
-            db_btn = gr.Button("Generate vector database...")
-            qachain_btn = gr.Button("Initialize question-answering chain...")
-            submit_btn = gr.Button("Submit")
-            clear_btn = gr.ClearButton([msg, chatbot])
+                submit_btn = gr.Button("Submit")
+                clear_btn = gr.ClearButton([msg, chatbot])
             
         # Preprocessing events
         #upload_btn.upload(upload_file, inputs=[upload_btn], outputs=[document])