Deploy DeepSeek LLM chatbotV2

app.py

@@ -5,46 +5,147 @@ import torch
 # Load DeepSeek LLM
 model_name = "deepseek-ai/deepseek-llm-7b-chat"  
 tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map="auto")
+# model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map="auto")
+
+model = AutoModelForCausalLM.from_pretrained(
+    model_name, 
+    torch_dtype=torch.float16, 
+    device_map="auto", 
+    offload_folder="offload_weights"
+)
+'''
+ValueError: The current device_map had weights offloaded to the disk. Please provide an offload_folder f
+or them. Alternatively, make sure you have safetensors installed if the model you are using offers the 
+weights in this format.
+
+'''
 
 SYSTEM_PROMPT = "You are a helpful AI assistant. Keep responses concise and informative."
 
+
+import wikipediaapi
+
+wiki_wiki = wikipediaapi.Wikipedia('en')
+
+def fetch_wikipedia(query):
+    """Fetch a summary from Wikipedia"""
+    page = wiki_wiki.page(query)
+    if page.exists():
+        return page.summary[:500]  # Limit to 500 chars
+    return "I couldn't find relevant Wikipedia information on that topic."
+
 def generate_response(message, history):
     history = history or []
     history.append(("User", message))
-    
-    # Add system message for better guidance
-    input_text = f"[SYSTEM] {SYSTEM_PROMPT}\n" + tokenizer.apply_chat_template(history, tokenize=False, add_generation_prompt=True)
+
+    # Check if the user asks for factual info
+    if "wikipedia" in message.lower():
+        query = message.lower().replace("wikipedia", "").strip()
+        wiki_info = fetch_wikipedia(query)
+        history.append(("Bot", wiki_info))
+        return history, ""
+
+    # Default chatbot behavior
+    chat_history = ""
+    for user, bot in history[-5:]:
+        chat_history += f"User: {user}\nBot: {bot}\n"
+
+    input_text = f"[SYSTEM] {SYSTEM_PROMPT}\n{chat_history}User: {message}\nBot:"
     inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
 
     streamer = tokenizer.streamer()
     model.generate(**inputs, streamer=streamer, max_length=512, pad_token_id=tokenizer.eos_token_id)
-    
+
     bot_message = ""
     for token in streamer:
         bot_message += token
         yield bot_message  
 
-# Create Gradio Chatbot UI with streaming
+    history.append(("Bot", bot_message))
+    return history, ""
+
+
+# # Function to handle chat with memory
+# def generate_response(message, history):
+#     history = history or []
+#     history.append(("User", message))
+
+#     # Format history for the model
+#     chat_history = ""
+#     for user, bot in history[-5:]:  # Limit history to last 5 exchanges to avoid exceeding token limit
+#         chat_history += f"User: {user}\nBot: {bot}\n"
+
+#     input_text = f"[SYSTEM] {SYSTEM_PROMPT}\n{chat_history}User: {message}\nBot:"
+#     inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
+
+#     streamer = tokenizer.streamer()
+#     model.generate(**inputs, streamer=streamer, max_length=512, pad_token_id=tokenizer.eos_token_id)
+
+#     bot_message = ""
+#     for token in streamer:
+#         bot_message += token
+#         yield bot_message  
+
+#     history.append(("Bot", bot_message))
+#     return history, ""
+
+# def generate_response(message, history):
+#     history = history or []
+#     history.append(("User", message))
+    
+#     # Add system message for better guidance
+#     input_text = f"[SYSTEM] {SYSTEM_PROMPT}\n" + tokenizer.apply_chat_template(history, tokenize=False, add_generation_prompt=True)
+#     inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
+
+#     streamer = tokenizer.streamer()
+#     model.generate(**inputs, streamer=streamer, max_length=512, pad_token_id=tokenizer.eos_token_id)
+    
+#     bot_message = ""
+#     for token in streamer:
+#         bot_message += token
+#         yield bot_message  
+
+
 with gr.Blocks() as demo:
-    gr.Markdown("### 🚀 DeepSeek LLM Chatbot (Streaming & Improved UI)")
+    gr.Markdown("# 🚀 DeepSeek LLM Chatbot with Memory & Wikipedia API")
     
     chatbot = gr.Chatbot()
-    msg = gr.Textbox(placeholder="Type your message here...", label="Your Message")
+    msg = gr.Textbox(placeholder="Ask me anything...", label="Your Message")
     clear_btn = gr.Button("Clear Chat")
     
     def respond(message, history):
-        history = history or []
-        bot_response = generate_response(message, history)
-        return bot_response, history + [("User", message), ("Bot", bot_response)]
-    
+        history, bot_message = generate_response(message, history)
+        return history, bot_message
+
     msg.submit(respond, inputs=[msg, chatbot], outputs=[chatbot, msg])
     clear_btn.click(lambda: ([], ""), outputs=[chatbot, msg])
 
+
+
+# # Create Gradio Chatbot UI with streaming
+# with gr.Blocks() as demo:
+#     gr.Markdown("### 🚀 DeepSeek LLM Chatbot (Streaming & Improved UI)")
+    
+#     chatbot = gr.Chatbot()
+#     msg = gr.Textbox(placeholder="Type your message here...", label="Your Message")
+#     clear_btn = gr.Button("Clear Chat")
+    
+#     def respond(message, history):
+#         history = history or []
+#         bot_response = generate_response(message, history)
+#         return bot_response, history + [("User", message), ("Bot", bot_response)]
+    
+#     msg.submit(respond, inputs=[msg, chatbot], outputs=[chatbot, msg])
+#     clear_btn.click(lambda: ([], ""), outputs=[chatbot, msg])
+
+
 demo.launch()
 
 
 
+
+
+
 # with gr.Blocks() as demo:
 #     gr.Markdown("### 🚀 DeepSeek LLM Chatbot (Streaming Enabled)")
 #     chat = gr.ChatInterface(fn=generate_response)

requirements.txt

@@ -2,3 +2,4 @@ gradio
 transformers
 torch
 accelerate
+wikipedia-api