up

gradio_llm_example.py

@@ -1,67 +1,56 @@
 import gradio as gr
-import time
+import datetime
 import random
+import time
 import os
 import shutil
+import pandas as pd
 # How to RUN code ==> gradio gradio_llm_example.py 
 
 
-from langchain import HuggingFacePipeline
-def load_llm_model(model: str = "google/flan-t5-large") -> HuggingFacePipeline:
-    llm = HuggingFacePipeline.from_model_id(
-        model_id=model,
-        task="text2text-generation",
-        model_kwargs={"max_length": 1500, "load_in_8bit": True},
-    )
-    return llm
-
 # Define text and title information 
-title1 = "##  </br> </br> </br> 🤗💬 QA App"
+title1 = "##  🤗 About QA App"
 
-title2 = " ##  </br> </br> </br> Gradio QA Bot"
+title2 = " ##  💬 Chat with QA Bot"
 
-intro = """ Welcome! This is not just any bot, it's a special one equipped with state-of-the-art natural language processing capabilities, and ready to answer your queries.
+title3 = " ##  🔧 Toolbox "
 
+title4 = " ##  ⚙️ Parameters"
 
-    Ready to explore? Let's get started!
-  
-    
-    * Step 1: Upload a PDF document.
-    * Step 2: Type in a question related to your document's content.
-    * Step 3: Get your answer!
- 
-    
-    Push clear cache before uploading a new doc!
+title5 = " ##  📊 Evaluation"
 
+intro = """ Welcome! This is not just any bot, it's a special one equipped with state-of-the-art natural language processing capabilities, and ready to answer your queries with/without the support of some additional documents.
+
+
+    **Ready to explore? Let's get started!** 🏁🎉 
+  
     
-    """
-
-about = """ 
-    ## </br> About
-    This app is an LLM-powered chatbot built using:
-    - [Streamlit](<https://streamlit.io/>)
-    - [HugChat](<https://github.com/Soulter/hugging-chat-api>)
-    - Chat Model = llama2-chat-hf 7B 
-    - Retreiver model = all-MiniLM-L6-v2
-
-    </br> 
-    💡 Note: No API key required!
-    
-     </br> 
-    Made with ❤️ by us
-     """
+    * Chat with the bot by typing any question you want 
+    and get your answers !
+    * You can load and select one or more documents to reinforce the bot's knowledge. 
+    Don't forget to validate and update your selection according to your choices.
+    * You can customize your model by selecting advanced options in the toolbox."""
 
+final_info = """ Made with ❤️ by us 🚀"""
 
 # Define theme ==> see gr.themes.builder()
 theme = gr.themes.Soft(
     primary_hue="emerald",
-    secondary_hue="emerald",
     neutral_hue="slate",
+    text_size=gr.themes.sizes.text_md,
 ).set(
-    body_background_fill_dark='*primary_50',
-    shadow_drop='*shadow_spread',
-    button_border_width='*block_border_width',
-    button_border_width_dark='*block_label_border_width'
+    body_text_color='*secondary_900',
+    body_text_size='*text_lg',
+    body_text_weight='500',
+    border_color_accent='*secondary_950',
+    link_text_color='*secondary_300',
+    block_border_color='*neutral_200',
+    block_border_width='*block_label_border_width',
+    block_label_background_fill='*primary_200',
+    block_title_text_color='*primary_350',
+    checkbox_border_color='*primary_300',
+    checkbox_border_color_selected_dark='*secondary_200',
+    button_primary_border_color='*primary_350'
 )
 
 
@@ -69,8 +58,8 @@ def upload_file(files_obj):
     """ Upload several files from drag and drop, and save them in local temp folder
     files_obj (type:list) : list of tempfile._TemporaryFileWrapper 
     return checkbox to display uploaded documents """
-    # Create local copy
     temp_file_path = "./temp"
+    # Create local copy
     if not os.path.exists(temp_file_path):
         os.makedirs(temp_file_path)
     # Save each file among list of given files
@@ -81,19 +70,21 @@ def upload_file(files_obj):
         shutil.copyfile(file_obj.name, os.path.join(temp_file_path, file_name))
     # return visible button for next selection 
     return {uploaded_check : gr.CheckboxGroup(choices=file_name_list, visible=True),
-            choose_btn : gr.Button(value="Choose", visible=True)}
+            choose_btn : gr.Button(visible=True),
+            clear_folder_btn : gr.Button(visible=True)}
+
+
 
 
-def read_content(content, files_name):
+def read_PDFcontent(content, files_name):
     """ Read and update the content variable (state) according to the several files_names to read from temp folder
     return updated content_var (type : list of str)
     return visible error_box to display logs error """
     content_list = list()
     text_list = list()
     # Parse one or several docs among the selected ones
+    temp_file_path = "./temp"
     for file_name in files_name :
-        print(file_name, type(file_name))
-        temp_file_path = "./temp"
         file_path = os.path.join(temp_file_path, file_name)
         # Read doc
         with open(file_path, "rb") as file:
@@ -102,28 +93,73 @@ def read_content(content, files_name):
                 #### YOUR FONCTION FOR CONTENT ==> must be str
                 my_content = str(content[:10])
                 content_list.append(my_content) 
-                text_list.append(f"File {file_name} ready to be used. \n")
+                text_list.append(f" {file_name} : ready ✅ \n ")
+                # print(content)
             except Exception as e:
                 print(f"Error occurred while writing the file: {e}")
-                text_list.append(f"Error occurred while writing the file {file_name}: {e}")
+                text_list.append(f" {file_name} : error ❌ \n")
     return {content_var : content_list,
-            error_box : gr.Textbox(value=f"""{" and ".join(text_list)} \n You can ask a question about the uploaded PDF document.""", visible=True)}
+            error_box : gr.Textbox(value=f"""{"".join(text_list)} """, visible=True)}
+
+
+def clear_folder():
+    temp_file_path = "./temp"
+    shutil.rmtree(temp_file_path)
+    return {uploaded_check : gr.CheckboxGroup(choices=[], visible=False), 
+            error_box :  gr.Textbox("", visible=False),
+            choose_btn : gr.Button(visible=False),
+            clear_folder_btn : gr.Button(visible=False),
+            context_box : gr.Textbox(""),
+            content_var : []}
+
+# def write_content (chat_history, download_counter):
+#     temp_file_path = "./Download_chat"
+#     if not os.path.exists(temp_file_path):
+#         os.makedirs(temp_file_path)
+#     file_name = str(download_counter)+"chat_Conversation.txt"
+#     file_path = os.path.join(temp_file_path, file_name)
+#     # write doc
+#     with open(file_path, "w") as file:
+#         for query_answer in chat_history :
+#             file.write(" \n ".join(query_answer))
+#             file.write(" \n ")
+#     new_count = int(download_counter)+1
+#     return { download_counter : gr.Number(new_count, visible=False)}
 
 
 ### YOUR model using the same inputand returning output 
 def my_model(message, chat_history, content_var,
             language_choice, model_choice, max_length, temperature,
             num_return_sequences, top_p, no_repeat_ngram_size):
+        
         #No LLM here, just respond with a random pre-made message
         if content_var == []:
-            bot_message = f"Pas de context : {content_var}" + random.choice(["Tell me more about it", 
-                                            "Cool, but I'm not interested", 
-                                            "Hmmmm, ok then"]) 
+            bot_message = f"Pas de contexte : {content_var}" +  "I'm not interested" 
         else:
-            bot_message = f" Voici le context {content_var}"
+            bot_message = f" Voici le contexte :  {content_var}"
         chat_history.append((message, bot_message))
+        
+        return "", chat_history, {context_box : gr.Textbox(visible=True, value=f'{"and".join(content_var)}')}   
 
-        return "", chat_history
+
+def evaluate_my_model(answer):
+    # true_answer.update("")
+    return {true_answer : gr.Textbox.update("") ,
+            updated_check : gr.Button(f"Model updated ✅ at : {datetime.datetime.now().strftime('%H:%M:%S')} " , 
+                                      visible = True)}
+
+    
+
+def display_my_metrics(metric_csv="./metrics.csv"):
+    df = pd.read_csv(metric_csv)
+    # df_metrics[0]=["Model", "P", "R", "F1", "⏳"]
+    return {df_metrics : gr.DataFrame(df, visible=True)}
+
+
+def clear_all():
+    return {msg: gr.Textbox("") , 
+            chatbot: gr.Chatbot(""),
+            context_box: gr.Textbox("") }
 
 
 def queue_bot(history):
@@ -132,71 +168,134 @@ def queue_bot(history):
     history[-1][1] = ""
     for character in bot_message:
         history[-1][1] += character
-        time.sleep(0.05)
+        time.sleep(0.005)
         yield history
 
 
+
+# Params
+temp_file_path = "./temp"
+
+
 # App
-with gr.Blocks(theme=gr.themes.Soft()) as gradioApp:
+with gr.Blocks(theme=theme) as gradioApp:
 
     # Initialize the document context variable as empty without any drag and drop
     content_var = gr.State([])
+    download_counter = gr.Number(0, visible=False)
+
 
     # Layout
-    with gr.Row():
-        # Row 1 : About
-        with gr.Column(scale=1, min_width=100):
-            # gr.Image("./logo_neovision.png")
-            logo_gr = gr.Markdown(""" <img src="file/logo_neovision.png" alt="logo" style="width:400px;"/>""") 
-            about_gr = gr.Markdown(about)   
-        
-        # Row 2 : Param
-        with gr.Column(scale=2, min_width=500):
-            title1_gr= gr.Markdown(title1)
-            intro_gr = gr.Markdown(intro)
-
-            #  Upload several documents
-            upload_button = gr.UploadButton("Browse files", label="Drag and drop your documents here", 
-                                            size="lg", scale=0, min_width=100,
-                                            file_types=["pdf"], file_count="multiple")
-            # invisible button while no documents uploaded
-            uploaded_check = gr.CheckboxGroup(label="Uploaded documents", visible=False, 
-                                      info="Do you want to use a supporting document?")
-            choose_btn = gr.Button(value="Choose", visible=False)
-            # uploading one or several docs and display other buttons
-            upload_button.upload(upload_file, upload_button, [uploaded_check, choose_btn])
-
-            # Read only one document
-            error_box = gr.Textbox(label="Reading files... ", visible=False) # display only when ready or error
-            choose_btn.click(read_content, inputs=[content_var, uploaded_check], outputs=[content_var, error_box])
-
-            # Select advanced options, to be given as input for your model
-            gr.Markdown(""" ## Toolbox """)
-            with gr.Accordion(label="Select advanced options",open=False):
-                model_choice = gr.Dropdown(["LLM", "Other"], label="Model", info="Choose your AI model")
-                language_choice = gr.Dropdown(["English", "French"], label="Language", info="Choose your language")
-                max_length = gr.Slider(label="Token length", minimum=1, maximum=100, value=50, step=1)
-                temperature= gr.Slider(label="Temperature", minimum=0.1, maximum=1, value=0.8, step=0.1)
-                num_return_sequences= gr.Slider(label="Return Sequence", minimum=0.1, maximum=50, value=1, step=0.1)
-                top_p= gr.Slider(label="top p", minimum=0.1, maximum=1, value=0.8, step=0.1)
-                no_repeat_ngram_size= gr.Slider(label="repeat", minimum=0.1, maximum=1, value=3, step=0.1)
+    gr.Markdown(""" <img src="file/logo_neovision.png" alt="logo" style="width:350px;"/>""") 
+
+    with gr.Column():
+        gr.Markdown(title1)
+        gr.Markdown(intro)
+        # gr.Markdown(final_info) 
+
+
+    # Row 1 : Intro + Param
+    with gr.Row(equal_height=True):
 
+        # with gr.Column(min_width=80, scale = 0):
+        #     gr.Markdown("    ")
 
-        # Row 3 : Chat
-        with gr.Column(scale=2, min_width=600):
+        # Row 2 : Chat
+        with gr.Column(min_width= 300, scale = 3):
             title2_gr = gr.Markdown(title2)
-            chatbot = gr.Chatbot(label="Bot", height=500)
+            chatbot = gr.Chatbot(label="Bot", height=300)
             msg = gr.Textbox(label="User", placeholder="Ask any question.")
-            ### YOUR MODEL TO ADAPT
-            msg.submit(my_model,
-                       inputs=[msg, chatbot, content_var,
-                                language_choice, model_choice, max_length, temperature,
-                                num_return_sequences, top_p, no_repeat_ngram_size],
-                                outputs=[msg, chatbot]).then(queue_bot, chatbot, chatbot)
-            clear = gr.ClearButton(components=[msg, chatbot], value="Clear console")
+            with gr.Row():
+                with gr.Column():
+                    # clear = gr.ClearButton(components=[msg, chatbot, context_box], value="Clear console")
+                    clear = gr.Button( value="🗑️ Clear console")
+                    download = gr.Button( value="📩 Download chat")
+                    updated_download = gr.Button(f"Last download ✅ at : {datetime.datetime.now().strftime('%H:%M:%S')} " , 
+                                      visible = False)
+                    
+                with gr.Column():   
+                    upload_button = gr.UploadButton("📁 Browse files", label="Drag and drop your documents here", 
+                                                file_types=["pdf"], file_count="multiple")
+                    uploaded_check = gr.CheckboxGroup(label=" 📁 Uploaded documents", visible=False, 
+                                            info="Do you want to use a supporting document?")
+                    with gr.Row():
+                        choose_btn = gr.Button(value="🖱️ Choose docs", visible=False)
+                        clear_folder_btn = gr.Button(value="🗑️ Clear docs", visible=False)
+                    # upload_iface = gr.Interface(fn=upload_file, 
+                    #                     inputs=gr.File(file_count="multiple", file_types=["pdf"]), 
+                    #                     outputs=[uploaded_check], 
+                    #                     description="📁 Browse files", allow_flagging="never")
+                    error_box = gr.Textbox(label="Files state... ", visible=False) # display only when ready or error
+
+        # Row 3 : Toolbox
+        with gr.Column(min_width= 100, scale = 1):
+            gr.Markdown(title4)
+            with gr.Accordion(label="Select advanced options",open=False):
+                model_choice = gr.Dropdown(["LLM", "Other"], label="Model", info="Choose your AI model")
+                language_choice = gr.Dropdown(["English", "French"], label="Language", info="Choose your language")
+                max_length = gr.Slider(label="Token length", minimum=1, maximum=100, value=50, step=1)
+                temperature= gr.Slider(label="Temperature", minimum=0.1, maximum=1, value=0.8, step=0.5)
+                num_return_sequences= gr.Slider(label="Return Sequence", minimum=0.1, maximum=50, value=1, step=0.5)
+                top_p= gr.Slider(label="top p", minimum=0.1, maximum=1, value=0.8, step=0.5)
+                no_repeat_ngram_size= gr.Slider(label="repeat", minimum=0.1, maximum=1, value=3, step=0.5)
+
+        # Evaluate
+        with gr.Column(min_width= 100, scale = 1):
+            gr.Markdown(title5)
+            with gr.Accordion(label=" Evaluate the model",open=True):
+                context_box = gr.Textbox(label= "📚 Context", 
+                                         placeholder=" The used context" )
+                true_answer = gr.Textbox(lines = 2, label= "📝 Right Answer", 
+                                            placeholder="Give an example of right answer and update the model")
+                update_btn = gr.Button(value = "Update ⚠️ 🔄" )
+                updated_check = gr.Button(value= "", visible = False)
+                df_metrics = gr.DataFrame(row_count=(5,  "fixed"), col_count=(3, "fixed"), visible=False)
+
+
+
+
+    ###### Chatbot ######
+    # Flag
+    logger = gr.CSVLogger()
+    logger.setup([chatbot], "Download_flagged")
+    # hf_writer = gr.HuggingFaceDatasetSaver(HF_TOKEN, "crowdsourced-calculator-demo")
+    download.click(lambda *args: logger.flag(args), [chatbot], None, preprocess=False)
+    download.click(lambda  : {updated_download:gr.Button(visible=True)}, None, updated_download, preprocess=False)
+
+    # YOUR MODEL TO ADAPT
+    msg.submit(my_model, inputs=[msg, chatbot, content_var,
+                    language_choice, model_choice, max_length, temperature,
+                    num_return_sequences, top_p, no_repeat_ngram_size],
+                    outputs=[msg, chatbot, context_box]).then(queue_bot, chatbot, chatbot)
+    # Chatbot clear
+    clear.click(clear_all, inputs=[] , outputs=[msg,  chatbot, context_box])
+    # Chatbot examples
+    example = gr.Examples(examples = [  ["What are the payment terms?"],
+                                    ["Do I become the owner of the developments made?"],
+                                    [" Can Neovision use a subcontractor and if so, under what conditions?"],
+                                    ["What are the termination conditions?"]],
+                        fn = my_model, inputs=[msg, chatbot, content_var,
+                    language_choice, model_choice, max_length, temperature,
+                    num_return_sequences, top_p, no_repeat_ngram_size],
+                    outputs=[msg, chatbot, context_box])
+
+
+    ###### DOCS ######
+    # uploading one or several docs and display other buttons
+    upload_button.upload(upload_file, [upload_button], [uploaded_check, choose_btn, clear_folder_btn])
+    # Read only one document
+    choose_btn.click(read_PDFcontent, inputs=[content_var, uploaded_check], outputs=[content_var, error_box])
+    # # clear
+    clear_folder_btn.click(clear_folder, inputs=[] , 
+                           outputs=[uploaded_check, error_box, choose_btn, clear_folder_btn, 
+                                    context_box, content_var])
 
+    
+    # evaluate and update model
+    update_btn.click(evaluate_my_model, inputs=true_answer, outputs=[true_answer, updated_check])
+    update_btn.click(display_my_metrics, inputs=None, outputs=df_metrics)
 
 gr.close_all()
 gradioApp.queue()
-gradioApp.launch(share=True, auth=("neovision", "gradio2023"))
+gradioApp.launch(share=True)
 #auth=("neovision", "gradio2023") to be placed inside the launch parameters

metrics.csv

@@ -0,0 +1,5 @@
+modèles,M1,M2
+P,895,905
+R,891,878
+F1,893,891
+Time (s),28.04,"32,67"