feat: switch to deepseek model for token-free operation

README.md

@@ -1,27 +1,27 @@
 ---
-title: AI Chat Assistant
-emoji: 💭
+title: DeepSeek R1 Chat
+emoji: 🧠
 colorFrom: blue
 colorTo: purple
 sdk: streamlit
 sdk_version: 1.41.1
 app_file: app.py
 pinned: false
-short_description: ChatGPT-like interface using Ollama
+short_description: Advanced Chat using DeepSeek-R1-Distill-8B
 ---
 
-# AI Chat Assistant
+# DeepSeek R1 Chat Assistant
 
-Simple and efficient chat interface powered by Ollama DeepSeek model.
+Powerful chat interface powered by DeepSeek-R1-Distill-Llama-8B model.
 
 ## Features
-- Clean chat interface
-- Message history
-- Streaming responses
-- Context awareness
-- Local inference
+- Advanced language understanding
+- Context-aware responses
+- Efficient 8B parameter model
+- Local CPU inference
+- Memory optimized
 
 ## Usage
 - Type your message and press Enter
 - Clear chat history using sidebar button
-- Maintains conversation context
+- Best for complex conversations

app.py

@@ -3,27 +3,35 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
 import logging
 
-# Configure page and logging
-st.set_page_config(page_title="AI Chat Assistant", page_icon="💭", layout="wide")
+# Configure page
+st.set_page_config(
+    page_title="DeepSeek R1 Chat",
+    page_icon="🧠",
+    layout="wide",
+    initial_sidebar_state="expanded"
+)
+
+# Set up logging and style
 logging.basicConfig(level=logging.INFO)
 
-# Custom CSS for chat interface
 st.markdown("""
 <style>
     .stChat { padding: 20px; border-radius: 10px; }
     .user-message { background-color: #e6f3ff; }
     .assistant-message { background-color: #f0f2f6; }
+    .stButton button { background-color: #2E86C1; }
 </style>
 """, unsafe_allow_html=True)
 
 @st.cache_resource
 def load_model():
-    model_name = "deepseek-ai/deepseek-chat-1.3b-base"  # Smaller DeepSeek model
+    model_name = "deepseek-ai/DeepSeek-R1-Distill-Llama-8B"
     
     try:
         tokenizer = AutoTokenizer.from_pretrained(
             model_name,
-            trust_remote_code=True
+            trust_remote_code=True,
+            padding_side='left'
         )
         tokenizer.pad_token = tokenizer.eos_token
         
@@ -31,9 +39,12 @@ def load_model():
             model_name,
             torch_dtype=torch.float32,
             low_cpu_mem_usage=True,
-            trust_remote_code=True
-        ).to("cpu")
+            trust_remote_code=True,
+            device_map='cpu'
+        )
         
+        model.eval()
+        torch.set_num_threads(8)
         return model, tokenizer
         
     except Exception as e:
@@ -42,30 +53,52 @@ def load_model():
 
 def generate_response(prompt, model, tokenizer):
     try:
-        # Format prompt for DeepSeek chat
-        chat_prompt = f"Human: {prompt}\n\nAssistant: Let me help you with that."
+        chat_prompt = f"""user
+{prompt}
+assistant
+I'll help you with that."""
+        
+        inputs = tokenizer(
+            chat_prompt,
+            return_tensors="pt",
+            padding=True,
+            truncation=True,
+            max_length=2048
+        )
         
-        inputs = tokenizer(chat_prompt, return_tensors="pt", padding=True)
+        # Create placeholder for streaming output
         message_placeholder = st.empty()
-        response_text = ""
+        full_response = ""
         
         with torch.inference_mode():
-            outputs = model.generate(
-                inputs["input_ids"],
-                max_length=512,
-                temperature=0.7,
-                top_p=0.95,
-                do_sample=True,
-                pad_token_id=tokenizer.eos_token_id,
-                attention_mask=inputs["attention_mask"],
-                num_return_sequences=1,
-                repetition_penalty=1.2
-            )
+            generated_ids = []
+            for i in range(512):  # Max new tokens
+                # Generate next token
+                outputs = model.generate(
+                    inputs["input_ids"] if not generated_ids else torch.cat([inputs["input_ids"], torch.tensor([generated_ids]).to(model.device)], dim=1),
+                    max_new_tokens=1,
+                    temperature=0.7,
+                    do_sample=True,
+                    top_p=0.95,
+                    repetition_penalty=1.1,
+                    pad_token_id=tokenizer.eos_token_id
+                )
+                
+                next_token = outputs[0][-1].item()
+                generated_ids.append(next_token)
+                
+                # Decode and display current state
+                current_output = tokenizer.decode(generated_ids, skip_special_tokens=True)
+                full_response = current_output
+                message_placeholder.markdown(full_response)
+                
+                # Check for end of generation
+                if next_token == tokenizer.eos_token_id:
+                    break
             
-            response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-            # Extract assistant's response
-            response = response.split("Assistant:")[-1].strip()
-            return response
+            # Clean up response
+            response = full_response.split("assistant")[-1].strip()
+            return response.split("user")[0].strip()
             
     except Exception as e:
         st.error(f"Error: {str(e)}")
@@ -77,29 +110,24 @@ def init_chat():
         st.session_state.model, st.session_state.tokenizer = load_model()
 
 def main():
-    st.title("💭 AI Chat Assistant")
+    st.title("🧠 DeepSeek R1 Chat Assistant")
     init_chat()
     
-    # Sidebar with controls
     with st.sidebar:
-        st.markdown("### Chat Controls")
-        if st.button("🗑️ Clear Chat", use_container_width=True):
+        st.markdown("### Chat Settings")
+        if st.button("🗑️ Clear History", use_container_width=True):
             st.session_state.messages = []
             st.rerun()
     
-    # Display chat messages
     for message in st.session_state.messages:
         with st.chat_message(message["role"]):
             st.markdown(message["content"])
     
-    # Chat input
-    if prompt := st.chat_input("Send a message..."):
-        # Add user message
+    if prompt := st.chat_input("Ask me anything..."):
         st.session_state.messages.append({"role": "user", "content": prompt})
         with st.chat_message("user"):
             st.markdown(prompt)
         
-        # Generate and display assistant response
         with st.chat_message("assistant"):
             context = "\n".join([
                 f"{m['role']}: {m['content']}" 
@@ -113,6 +141,7 @@ def main():
             )
             
             if response:
+                st.markdown(response)
                 st.session_state.messages.append(
                     {"role": "assistant", "content": response}
                 )