chatbot_ui v2

app.py

@@ -1,5 +1,47 @@
 import gradio as gr
 import speech_recognition as sr
+from datetime import datetime, timedelta
+import os
+import threading
+from huggingface_hub import InferenceClient
+from dotenv import load_dotenv
+load_dotenv()
+
+
+system_prompt = """
+You are an AI coding assistant designed to solve coding problems and provide code snippets based on the user's query. When given a query, follow these guidelines.
+1. Return only the necessary and helpful code.
+2. Include any related details that enhance understanding or usability of the code.
+3. Ensure the code is clean, efficient, and follows best practices.
+4. Add comments to explain complex or non-obvious parts of the code.
+5. If there are multiple possible solutions, provide the most optimal one first.
+"""
+ip_requests = {}
+ip_requests_lock = threading.Lock()
+
+def allow_ip(request: gr.Request, show_error=True):
+    ip = request.headers.get("X-Forwarded-For")
+    now = datetime.now()
+    window = timedelta(hours=24)
+    with ip_requests_lock:
+        if ip in ip_requests:
+            ip_requests[ip] = [timestamp for timestamp in ip_requests[ip] if now - timestamp < window]
+        if len(ip_requests.get(ip, [])) >= 15:
+            raise gr.Error("Rate limit exceeded. Please try again tomorrow or use your Hugging Face Pro token.", visible=show_error)
+        ip_requests.setdefault(ip, []).append(now)
+        print("ip_requests", ip_requests)
+    return True
+
+def inference(prompt, hf_token, model, model_name, max_new_tokens):
+    messages = [{"role": "system", "content": system_prompt}, {"role": "user", "content": prompt}]
+    if hf_token is None or not hf_token.strip():
+        hf_token = os.getenv("HF_TOKEN")
+    client = InferenceClient(model=model, token=hf_token)
+    tokens = f"**`{model_name}`**\n\n"
+    for completion in client.chat_completion(messages, max_tokens=max_new_tokens, stream=True):
+        token = completion.choices[0].delta.content
+        tokens += token
+    yield tokens
 
 def speech_to_text(audio):
     recognizer = sr.Recognizer()
@@ -16,105 +58,91 @@ def speech_to_text(audio):
         text = f"Could not process the audio, please try to record one more time"
     return text
 
-def chatbot_response(user_input):
-    if user_input.lower() == "hello":
-        return "Hi there! How can I help you?"
-    elif user_input.lower() == "how are you?":
-        return "I'm just a bot, but I'm functioning well! How about you?"
-    elif user_input.lower() == "what can you do?":
-        return "I can help answer questions, provide information, and assist with various tasks."
-    else:
-        return "I'm not sure how to respond to that. Can you please ask something else?"
+def respond(message, chat_history, system_prompt, hf_token, model_id, max_new_tokens):
+    bot_message = ""
+    for token in inference(message, hf_token, model_id, model_id.split("/")[-1], max_new_tokens):
+        bot_message += token
+    chat_history.append((message, bot_message))
+    yield "", chat_history
+        
+def clear_chat(chat_history):
+    return []
+
+def transcribe_audio(audio):
+    text = speech_to_text(audio)
+    return text
 
 def llm_ui():
     with gr.Blocks() as demo:
-        gr.Markdown("# AI Chatbot")
-
-        def respond(message, chat_history):
-            bot_message = chatbot_response(message)
-            chat_history.append((message, bot_message))
-            return "", chat_history
-
-        def clear_chat(chat_history):
-            return []
-
-        def load_previous_conversation():
-            # Logic to load previous conversation
-            return []
-
-        def transcribe_audio(audio):
-            text = speech_to_text(audio)
-            return text
+        
+        model_ids_llm = ["Mistralai/Mistral-7B-Instruct-v0.2", "meta-llama/Meta-Llama-3-8B-Instruct",  "meta-llama/Meta-Llama-3.1-8B-Instruct"]
+        
+        gr.Markdown("# AI Coding Assistant")
 
         with gr.Row():
             chatbot = gr.Chatbot()
+            
+        hf_token_box = gr.Textbox(lines=1, placeholder="Your Hugging Face token - Check if you have access to selected model", label="Hugging Face Token", type="password")
+            
         with gr.Row():
-            model_selection = gr.Dropdown(["Model 1", "Model 2", "Model 3"], label="Select Model", scale=3)
+            model_selection = gr.Dropdown(choices=model_ids_llm, value=model_ids_llm[0], label="Model", scale=3)
             clear_button = gr.Button("Clear Chat", scale=2)
-            load_button = gr.Button("Load Previous Conversation", scale=2)
-        with gr.Row():
-            user_input = gr.Textbox(placeholder="Type your message here...", label="User Input", show_label=False, scale=8)
-            send_button = gr.Button("Send", scale=2, elem_id="send-button")
+        
+        with gr.Group():
+            with gr.Row():
+                user_input = gr.Textbox(placeholder="Type your coding problem here...", label="User Input", show_label=False, scale=8)
+                send_button = gr.Button("Send", scale=2, variant = "primary")
         with gr.Row():
             voice_input = gr.Microphone(type="filepath", label="Voice Input", scale=7)
             voice_button = gr.Button("Use Audio as User Input", scale=3)
             voice_button.click(transcribe_audio, inputs=voice_input, outputs=user_input)
 
-        send_button.click(respond, [user_input, chatbot], [user_input, chatbot])
-        clear_button.click(clear_chat, [chatbot], [chatbot])
-        load_button.click(load_previous_conversation, [], [chatbot])
+        max_new_tokens_slider = gr.Slider(minimum=50, maximum=2000, value=500, step=10, label="Max New Tokens", info="Maximum number of tokens to generate in the response.")
 
-        demo.css = """
-        #send-button {
-            background-color: orange;
-        }
-        """
+        # Wrap system_prompt in a Gradio component
+        system_prompt_component = gr.State(value=system_prompt)
+
+        send_button.click(respond, [user_input, chatbot, system_prompt_component, hf_token_box, model_selection, max_new_tokens_slider], [user_input, chatbot], scroll_to_output=True)
+        clear_button.click(clear_chat, [chatbot], [chatbot])
     return demo
 
 def multimodal_llm_ui():
     with gr.Blocks() as demo:
-        gr.Markdown("# AI Chatbot")
-
-        def respond(message, chat_history):
-            bot_message = chatbot_response(message)
-            chat_history.append((message, bot_message))
-            return "", chat_history
-
-        def clear_chat(chat_history):
-            return []
-
-        def load_previous_conversation():
-            # Logic to load previous conversation
-            return []
-
-        def transcribe_audio(audio):
-            text = speech_to_text(audio)
-            return text
+        
+        model_ids_multimodal = ["Model-1", "Model-2",  "Model-3"]
+        
+        gr.Markdown("# Coding Vision Model")
 
         with gr.Row():
             chatbot = gr.Chatbot(height=550)
             with gr.Column():
                 with gr.Row():
-                    model_selection = gr.Dropdown(["Model 1", "Model 2", "Model 3"], label="Select Model", scale=3)
+                    model_selection = gr.Dropdown(choices=model_ids_multimodal, value=model_ids_multimodal[0], label="Select Model", scale=3)
                     clear_button = gr.Button("Clear Chat", scale=2)
-                    load_button = gr.Button("Load Previous Conversation", scale=2)
-
+                
+                hf_token_box = gr.Textbox(lines=1, placeholder="Your Hugging Face token - Check if you have access to selected model", label="Hugging Face Token", type="password")
+                
                 image_input = gr.Image(type="filepath", label="Input your Image Here....")
-                with gr.Row():
-                    user_input = gr.Textbox(placeholder="Type your message here...", label="User Input", show_label=False, scale=8)
-                    send_button = gr.Button("Send", scale=2, elem_id="send-button")
+                
+                with gr.Group():
+                    with gr.Row():
+                        user_input = gr.Textbox(placeholder="Type your problem here...", label="User Input", show_label=False, scale=8)
+                        send_button = gr.Button("Send", scale=2, elem_id="send-button", variant = "primary")
 
                 with gr.Row():
                     voice_input = gr.Microphone(type="filepath", label="Voice Input", scale=7)
                     voice_button = gr.Button("Use Audio as User Input", scale=3)
                     voice_button.click(transcribe_audio, inputs=voice_input, outputs=user_input)
 
-        send_button.click(respond, [user_input, chatbot], [user_input, chatbot])
+                max_new_tokens_slider = gr.Slider(minimum=50, maximum=2000, value=500, step=10, label="Max New Tokens", info="Maximum number of tokens to generate in the response.")
+
+        # Wrap system_prompt in a Gradio component
+        system_prompt_component = gr.State(value=system_prompt)
+
+        send_button.click(respond, [user_input, chatbot, system_prompt_component, hf_token_box, model_selection, max_new_tokens_slider], [user_input, chatbot])
         clear_button.click(clear_chat, [chatbot], [chatbot])
-        load_button.click(load_previous_conversation, [], [chatbot])
     return demo
 
-demo = gr.TabbedInterface([llm_ui(), multimodal_llm_ui()], ["LLM", "Image + LLM"],
-                         theme='snehilsanyal/scikit-learn')
+demo = gr.TabbedInterface([llm_ui(), multimodal_llm_ui()], ["LLM", "Image + LLM"])
 
-demo.launch()
+demo.launch()

requirements.txt

@@ -1 +1,4 @@
-SpeechRecognition
+SpeechRecognition
+gradio
+dotenv
+huggingface_hub