fin-tuning

app.py

@@ -10,20 +10,18 @@ model = AutoModelForCausalLM.from_pretrained(
     batch_size=1  # Ensures efficient processing
 )
 
-# System Prompt for Logical Responses
-SYSTEM_PROMPT = """You are an AI assistant. Respond concisely, logically, and naturally. Avoid making up information. Maintain context."""
+# System Prompt for Direct Responses
+SYSTEM_PROMPT = """You are an AI assistant. Answer **directly** and **concisely**. Avoid unnecessary explanations or small talk."""
 
-# Chat function with improved logic
+# Chat function with direct responses
 def chat(message, history):
-    history_str = "\n".join([f"User: {entry[0]}\nAI: {entry[1]}" for entry in history[-5:]])  # Keep last 5 exchanges for context
+    history_str = "\n".join([f"User: {entry[0]}\nAI: {entry[1]}" for entry in history[-5:]])  # Keep last 5 exchanges
     prompt = f"{SYSTEM_PROMPT}\n\n{history_str}\nUser: {message}\nAI:"
     
-    response = ""
-    for text in model(prompt, stream=True, temperature=0.5, max_new_tokens=100):
-        response += text
-        yield response  # Stream response dynamically
+    response = model(prompt, temperature=0.3, max_new_tokens=50, stop=["\nUser:"])
+    return response.strip()
 
-# Gradio UI (unchanged)
+# Gradio UI
 iface = gr.ChatInterface(fn=chat, title="LLaMA 3 Chatbot", theme="compact")
 
 # Launch app