Update app.py

app.py

@@ -22,25 +22,16 @@ import tqdm
 import accelerate
 import re
 
-# Static PDF file link
-static_pdf_link = "https://huggingface.co/spaces/CCCDev/PDFChat/resolve/main/Data-privacy-policy.pdf"
-
-list_llm = ["mistralai/Mistral-7B-Instruct-v0.2", "mistralai/Mixtral-8x7B-Instruct-v0.1",
-            "mistralai/Mistral-7B-Instruct-v0.1", "google/gemma-7b-it", "google/gemma-2b-it",
-            "HuggingFaceH4/zephyr-7b-beta", "HuggingFaceH4/zephyr-7b-gemma-v0.1",
-            "meta-llama/Llama-2-7b-chat-hf", "microsoft/phi-2",
-            "TinyLlama/TinyLlama-1.1B-Chat-v1.0", "mosaicml/mpt-7b-instruct", "tiiuae/falcon-7b-instruct",
-            "google/flan-t5-xxl"]
+list_llm = ["mistralai/Mistral-7B-Instruct-v0.2"]
 list_llm_simple = [os.path.basename(llm) for llm in list_llm]
+pdf_url = "path/to/your/static.pdf"  # Replace with your static PDF URL or path
 
 
 # Load PDF document and create doc splits
-def load_doc(file_path, chunk_size, chunk_overlap):
-    loader = PyPDFLoader(file_path)
+def load_doc(pdf_url, chunk_size, chunk_overlap):
+    loader = PyPDFLoader(pdf_url)
     pages = loader.load()
-    text_splitter = RecursiveCharacterTextSplitter(
-        chunk_size=chunk_size,
-        chunk_overlap=chunk_overlap)
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)
     doc_splits = text_splitter.split_documents(pages)
     return doc_splits
 
@@ -60,44 +51,13 @@ def create_db(splits, collection_name):
 
 # Initialize langchain LLM chain
 def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
-    progress(0.1, desc="Initializing HF tokenizer...")
-
     progress(0.5, desc="Initializing HF Hub...")
-    if llm_model == "mistralai/Mixtral-8x7B-Instruct-v0.1":
-        llm = HuggingFaceEndpoint(
-            repo_id=llm_model,
-            temperature=temperature,
-            max_new_tokens=max_tokens,
-            top_k=top_k,
-            load_in_8bit=True,
-        )
-    elif llm_model in ["HuggingFaceH4/zephyr-7b-gemma-v0.1", "mosaicml/mpt-7b-instruct"]:
-        raise gr.Error("LLM model is too large to be loaded automatically on free inference endpoint")
-    elif llm_model == "microsoft/phi-2":
-        llm = HuggingFaceEndpoint(
-            repo_id=llm_model,
-            temperature=temperature,
-            max_new_tokens=max_tokens,
-            top_k=top_k,
-            trust_remote_code=True,
-            torch_dtype="auto",
-        )
-    elif llm_model == "TinyLlama/TinyLlama-1.1B-Chat-v1.0":
-        llm = HuggingFaceEndpoint(
-            repo_id=llm_model,
-            temperature=temperature,
-            max_new_tokens=250,
-            top_k=top_k,
-        )
-    elif llm_model == "meta-llama/Llama-2-7b-chat-hf":
-        raise gr.Error("Llama-2-7b-chat-hf model requires a Pro subscription...")
-    else:
-        llm = HuggingFaceEndpoint(
-            repo_id=llm_model,
-            temperature=temperature,
-            max_new_tokens=max_tokens,
-            top_k=top_k,
-        )
+    llm = HuggingFaceEndpoint(
+        repo_id=llm_model,
+        temperature=temperature,
+        max_new_tokens=max_tokens,
+        top_k=top_k,
+    )
 
     progress(0.75, desc="Defining buffer memory...")
     memory = ConversationBufferMemory(
@@ -132,18 +92,14 @@ def create_collection_name(filepath):
         collection_name = 'A' + collection_name[1:]
     if not collection_name[-1].isalnum():
         collection_name = collection_name[:-1] + 'Z'
-    print('Filepath: ', filepath)
-    print('Collection name: ', collection_name)
     return collection_name
 
 
 # Initialize database
-def initialize_database(chunk_size, chunk_overlap, progress=gr.Progress()):
-    file_path = static_pdf_link
-    progress(0.1, desc="Creating collection name...")
-    collection_name = create_collection_name(file_path)
+def initialize_database(pdf_url, chunk_size, chunk_overlap, progress=gr.Progress()):
+    collection_name = create_collection_name(pdf_url)
     progress(0.25, desc="Loading document...")
-    doc_splits = load_doc(file_path, chunk_size, chunk_overlap)
+    doc_splits = load_doc(pdf_url, chunk_size, chunk_overlap)
     progress(0.5, desc="Generating vector database...")
     vector_db = create_db(doc_splits, collection_name)
     progress(0.9, desc="Done!")
@@ -152,7 +108,6 @@ def initialize_database(chunk_size, chunk_overlap, progress=gr.Progress()):
 
 def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
     llm_name = list_llm[llm_option]
-    print("llm_name: ", llm_name)
     qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, top_k, vector_db, progress)
     return qa_chain, "Complete!"
 
@@ -178,7 +133,6 @@ def conversation(qa_chain, message, history):
     response_source1_page = response_sources[0].metadata["page"] + 1
     response_source2_page = response_sources[1].metadata["page"] + 1
     response_source3_page = response_sources[2].metadata["page"] + 1
-
     new_history = history + [(message, response_answer)]
     return qa_chain, gr.update(
         value=""), new_history, response_source1, response_source1_page, response_source2, response_source2_page, response_source3, response_source3_page
@@ -200,72 +154,57 @@ def demo():
             <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 a reply.
             """)
 
-        with gr.Tab("Step 2 - Process document"):
-            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():
-                    chunk_size = gr.Slider(64, 4096, value=512, step=32, label="Text chunk size",
-                                           info="Text length of each document chunk being embedded into the vector database. Default is 512.")
-                    chunk_overlap = gr.Slider(0, 1024, value=24, step=8, label="Text chunk overlap",
-                                              info="Text overlap between each document chunk being embedded into the vector database. Default is 24.")
+        with gr.Tab("Step 1 - Upload PDF"):
+            gr.Markdown("Using static PDF link: path/to/your/static.pdf")
 
-            initialize_db = gr.Button("Process document")
+        with gr.Tab("Step 2 - Process document"):
+            gr.Markdown("Processing document automatically...")
 
-            with gr.Row():
-                output_db = gr.Textbox(label="Database initialization steps", placeholder="", show_label=False)
-                with gr.Accordion("Vector database collection details", open=False):
-                    collection = gr.Textbox(label="Collection name", placeholder="", show_label=False)
+        with gr.Tab("Step 3 - Initialize QA chain"):
+            gr.Markdown("Initializing QA chain automatically...")
 
-        with gr.Tab("Step 3 - Initialize LLM"):
-            with gr.Row():
-                llm_options = gr.Dropdown(list_llm_simple, label="Choose open-source LLM",
-                                          value="Mistral-7B-Instruct-v0.2",
-                                          info="Choose among the proposed open-source LLMs")
-            with gr.Accordion("Advanced LLM options", open=False):
+        with gr.Tab("Step 4 - Chatbot"):
+            chatbot = gr.Chatbot(height=300)
+            with gr.Accordion("Advanced - Document references", open=False):
                 with gr.Row():
-                    llm_temperature = gr.Slider(0.01, 1.0, value=0.1, step=0.01, label="LLM temperature",
-                                                info="LLM sampling temperature, in [0.01,1.0] range. Default is 0.1")
-                    llm_max_tokens = gr.Slider(32, 1024, value=512, step=16, label="Max tokens",
-                                               info="Maximum number of new tokens to be generated, in [32,1024] range. Default is 512")
-                    llm_top_k = gr.Slider(1, 40, value=20, step=1, label="Top K",
-                                          info="The number of highest probability vocabulary tokens to keep for top-k-filtering. Default is 20.")
-
-            initialize_llm = gr.Button("Initialize LLM")
-
+                    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():
-                output_llm = gr.Textbox(label="LLM initialization steps", placeholder="", show_label=False)
-
-        with gr.Tab("Step 4 - Start chatting"):
-            chatbot = gr.Chatbot(label="PDF chatbot", height=500)
-            msg = gr.Textbox(label="Your question", placeholder="Type your question here...", show_label=False)
-            clear = gr.Button("Clear chat")
-
-            with gr.Accordion("Document sources (3)", open=False):
-                gr.Markdown("Source 1")
-                response_src1 = gr.Textbox(label="Source 1", placeholder="", show_label=False)
-                response_src1_page = gr.Number(label="Page number", value=0, precision=0, interactive=False)
-                gr.Markdown("Source 2")
-                response_src2 = gr.Textbox(label="Source 2", placeholder="", show_label=False)
-                response_src2_page = gr.Number(label="Page number", value=0, precision=0, interactive=False)
-                gr.Markdown("Source 3")
-                response_src3 = gr.Textbox(label="Source 3", placeholder="", show_label=False)
-                response_src3_page = gr.Number(label="Page number", value=0, precision=0, interactive=False)
-
-        initialize_db.click(initialize_database,
-                            inputs=[chunk_size, chunk_overlap],
-                            outputs=[vector_db, collection_name, output_db])
-        initialize_llm.click(initialize_LLM,
-                             inputs=[llm_options, llm_temperature, llm_max_tokens, llm_top_k, vector_db],
-                             outputs=[qa_chain, output_llm])
-        msg.submit(conversation,
-                   inputs=[qa_chain, msg, chatbot],
-                   outputs=[chatbot, msg, chatbot, response_src1, response_src1_page, response_src2, response_src2_page,
-                            response_src3, response_src3_page])
-        clear.click(lambda: None, None, chatbot, queue=False)
-        clear.click(lambda: None, None, msg, queue=False)
-
+                msg = gr.Textbox(placeholder="Type message (e.g. 'What is this document about?')", container=True)
+            with gr.Row():
+                submit_btn = gr.Button("Submit message")
+                clear_btn = gr.ClearButton([msg, chatbot], value="Clear conversation")
+
+        # Automatic preprocessing
+        db_progress = gr.Textbox(label="Vector database initialization", value="Initializing...")
+        db_btn = gr.Button("Generate vector database", visible=False)
+        qachain_btn = gr.Button("Initialize Question Answering chain", visible=False)
+        llm_progress = gr.Textbox(value="None", label="QA chain initialization")
+
+        def auto_initialize():
+            vector_db, collection_name, db_status = initialize_database(pdf_url, 512, 24)
+            qa_chain, llm_status = initialize_LLM(0, 0.1, 512, 20, vector_db)
+            return vector_db, collection_name, db_status, qa_chain, llm_status, "Initialization complete."
+
+        demo.load(auto_initialize, [], [vector_db, collection_name, db_progress, qa_chain, llm_progress])
+
+        # Chatbot events
+        msg.submit(conversation, \
+                   inputs=[qa_chain, msg, chatbot], \
+                   outputs=[qa_chain, msg, chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3,
+                            source3_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,
+                                  doc_source3, source3_page], \
+                         queue=False)
     return demo.queue().launch(debug=True)