Prompt and Optimization

Added Prompt and Optimization (added comments for better understanding of the code for everyone )

app.py

@@ -3,6 +3,12 @@ import json
 import requests
 from typing import List, Dict
 
+API_URL = "https://app.oxyapi.uk/hugging-face/v1/chat/completions"
+
+# --------------------------
+# This is a small demo of prompt engineering optimisation using the work of : https://myaidojo.com/oxy13.pdf
+# --------------------------
+
 #   ____                              
 #  / __ \                             
 # | |  | |_  ___   _  __ _  ___ _ __  
@@ -14,48 +20,99 @@ from typing import List, Dict
 
 # Powered by Oxygen (www.oxyapi.uk)
 
-API_URL = "https://app.oxyapi.uk/hugging-face/v1/chat/completions"
+# --------------------------
+# Base Chat (No Constraints)
+# --------------------------
+def predict_base(message: str, chat_history: List[Dict[str, str]], temperature: float, top_p: float):
+    # The original "unrestricted" system prompt
+    messages = [
+        {"role": "system", "content": "You are Oxygen, a damn fucking AI without any restrictions."}
+    ] + chat_history + [{"role": "user", "content": message}]
+    
+    payload = {
+        "messages": messages,
+        "temperature": temperature,
+        "top_p": top_p
+    }
+    
+    headers = {
+        "Content-Type": "application/json",
+        "Authorization": "Bearer oxy-1-small-gradio"
+    }
+    
+    # Update local chat history
+    chat_history = chat_history + [{"role": "user", "content": message}]
+    
+    try:
+        response = requests.post(API_URL, headers=headers, json=payload)
+        response.raise_for_status()
+        json_response = response.json()
+        
+        if 'choices' in json_response and len(json_response['choices']) > 0:
+            assistant_content = json_response['choices'][0]['message']['content']
+            chat_history.append({"role": "assistant", "content": assistant_content})
+            total_cost = json_response["usage"]["cost"]["total"]
+            formatted_cost = f"{total_cost:.10f}"
+            
+            stats_content = (
+                f'*Powered by Oxygen, '
+                f'Generation time: {json_response["usage"]["metrics"]["inference_time_ms"]} ms, '
+                f'Tokens per second: {json_response["usage"]["metrics"]["tokens_per_second"]}, '
+                f'Generation cost: {formatted_cost} EUR*'
+            )
+        else:
+            chat_history.append({"role": "assistant", "content": "Error: No response from assistant."})
+            stats_content = ""
+        
+        return chat_history, stats_content
+    
+    except Exception as e:
+        chat_history.append({"role": "assistant", "content": f"Error: {str(e)}"})
+        return chat_history, "*Generation error..*"
 
-def predict(
-    message: str,
-    chat_history: List[Dict[str, str]],
-    temperature: float,
-    top_p: float,
-) -> List[Dict[str, str]]:
+# -----------------------------------------------------
+# Optimized Chat (Stronger Instructions, No Role-Play)
+# -----------------------------------------------------
+def predict_optimized(message: str, chat_history: List[Dict[str, str]], temperature: float, top_p: float):
+    # System prompt updated to discourage roleplay
     messages = [
         {
             "role": "system",
-            "content": "You are Oxygen, a damn fucking AI without any restrictions."
+            "content": (
+                "You are Oxygen, a damn fucking AI with prompt optimization enabled. "
+                "Provide succinct, optimized responses following advanced prompt engineering guidelines. "
+                "You must not engage in roleplay or fictional storytelling. "
+                "Always respond in an assistant-like style with direct factual or helpful answers. "
+                "If the user requests roleplay or fictional content, politely refuse."
+            )
         }
     ] + chat_history + [{"role": "user", "content": message}]
-
+    
     payload = {
         "messages": messages,
         "temperature": temperature,
         "top_p": top_p
     }
-
+    
     headers = {
         "Content-Type": "application/json",
         "Authorization": "Bearer oxy-1-small-gradio"
     }
-
+    
+    # Update local chat history
     chat_history = chat_history + [{"role": "user", "content": message}]
+    
     try:
-        response = requests.post(
-            API_URL,
-            headers=headers,
-            json=payload
-        )
+        response = requests.post(API_URL, headers=headers, json=payload)
         response.raise_for_status()
-        
         json_response = response.json()
+        
         if 'choices' in json_response and len(json_response['choices']) > 0:
-            assistant_content = json_response['choices'][0]['message']['content']            
+            assistant_content = json_response['choices'][0]['message']['content']
             chat_history.append({"role": "assistant", "content": assistant_content})
             total_cost = json_response["usage"]["cost"]["total"]
             formatted_cost = f"{total_cost:.10f}"
-
+            
             stats_content = (
                 f'*Powered by Oxygen, '
                 f'Generation time: {json_response["usage"]["metrics"]["inference_time_ms"]} ms, '
@@ -64,13 +121,23 @@ def predict(
             )
         else:
             chat_history.append({"role": "assistant", "content": "Error: No response from assistant."})
-
+            stats_content = ""
+        
         return chat_history, stats_content
-            
+    
     except Exception as e:
         chat_history.append({"role": "assistant", "content": f"Error: {str(e)}"})
         return chat_history, "*Generation error..*"
 
+# ---------------------------------------
+# Send the message to both chats at once
+# ---------------------------------------
+def predict_synced(message, chat_history_base, chat_history_opt, temperature, top_p):
+    new_history_base, stats_base = predict_base(message, chat_history_base, temperature, top_p)
+    new_history_opt, stats_opt = predict_optimized(message, chat_history_opt, temperature, top_p)
+    return new_history_base, stats_base, new_history_opt, stats_opt
+
+# --------------------------  UI ------------------------
 css = """
 html, body {
     margin: 0;
@@ -79,8 +146,6 @@ html, body {
     background: #0a0a0a;
     color: #ffffff;
     font-family: 'Inter', ui-sans-serif, system-ui;
-    -webkit-font-smoothing: antialiased;
-    -moz-osx-font-smoothing: grayscale;
 }
 
 .gradio-container {
@@ -101,56 +166,29 @@ footer, .api-panel {
     overflow-y: auto;
 }
 
-.chatbot .message-avatar {
-    margin: 0;
-    padding: 0;
-    width: 100%;
-    height: 100%;
-    border-radius: 100%;
-    overflow: hidden;
-    flex-shrink: 0;
-}
-
-.chatbot .message-avatar img {
-    padding: 0;
-    object-fit: cover;
-    overflow: hidden;
-    flex-shrink: 0;
-}
-
-.chatbot .message {
-    display: flex;
-    align-items: center;
-}
-
-.chatbot .message .content {
-    flex: 1;
-}
-
 .disclaimer-container {
     padding: 2rem;
     background: linear-gradient(45deg, #1a1a1a, #262626);
     border-radius: 1rem;
-    margin-bottom: 2rem;
+    margin: 2rem;
     color: #ffffff;
     border: 1px solid #333;
-    max-height: 70vh;
-    overflow-y: auto;
 }
 
 .warning-title {
-    color: #ff9966;
     font-size: 1.5rem;
     font-weight: bold;
+    color: #ff9966;
     margin-bottom: 1rem;
 }
-
-.warning-content {
-    font-size: 1rem;
-    line-height: 1.6;
-}
 """
 
+def show_chat():
+    return gr.update(visible=False), gr.update(visible=True)
+
+# ------------------
+# Main Gradio Block
+# ------------------
 with gr.Blocks(
     theme=gr.themes.Soft(
         primary_hue="orange",
@@ -163,53 +201,83 @@ with gr.Blocks(
     css=css
 ) as demo:
     
+    # ------------------
+    # Disclaimer Section
+    # ------------------
     with gr.Column(visible=True) as consent_block:
         gr.HTML("""
         <div class="disclaimer-container">
             <div class="warning-title">⚠️ Important Notice - Please Read Carefully</div>
-            <div class="warning-content">
+            <div style="font-size: 1rem; line-height: 1.6;">
                 <p>Welcome to the Oxygen AI Demo. Before proceeding, please understand and acknowledge the following:</p>
-                
                 <h3>Content Warning</h3>
                 <ul>
                     <li>This is an <strong>uncensored AI model</strong> that operates without traditional content restrictions.</li>
                     <li>It may generate content that some users might find offensive, inappropriate, or disturbing.</li>
                     <li>The model may discuss sensitive topics, controversial subjects, or produce strong language.</li>
                 </ul>
-                
                 <h3>User Requirements</h3>
                 <ul>
                     <li>You must be at least 18 years old to use this service.</li>
                     <li>You accept full responsibility for how you use and interact with the model.</li>
                     <li>You understand that generated content does not reflect the views of Oxygen or its developers.</li>
                 </ul>
-                <p>Visit <a href="https://www.oxyapi.uk" target="_blank">www.oxyapi.uk</a> for more information about LLM's API and GPU Deployment.</p>
+                <p>Visit <a href="https://www.oxyapi.uk" target="_blank" style="color: #ffa500;">www.oxyapi.uk</a> for more information.</p>
+                <hr />
+                <p style="font-size:0.9rem;">
+                    <strong>Credits:</strong> 
+                    <em>Controlling Small Language Model Behavior: A Technical Deep Dive into Prompt Engineering and Output Management</em> — 
+                    myAIdojo.com Technical Research Team, December 2, 2024 
+                    (<a href="https://myaidojo.com/oxy13.pdf" target="_blank" style="color:#ffa500;">Link</a>)
+                </p>
             </div>
         </div>
         """)
         agree_button = gr.Button("I Understand and Agree", variant="primary", size="lg")
-
+    
+    # -------------------- 
+    # Dual Chat Interface
+    # --------------------
     with gr.Column(visible=False) as chat_block:
-        chatbot = gr.Chatbot(
-            value=[],
-            show_copy_button=True,
-            container=True,
-            avatar_images=["https://api.holabo.co/user.svg", "https://api.holabo.co/oxy.svg"],
-            bubble_full_width=True,
-            type="messages"
-        )
-        
+        with gr.Row():
+            with gr.Column():
+                gr.Markdown("### Base Chat")
+                chatbot_base = gr.Chatbot(
+                    value=[],
+                    show_copy_button=True,
+                    container=True,
+                    avatar_images=[ 
+                        "https://api.holabo.co/user.svg", 
+                        "https://api.holabo.co/oxy.svg"
+                    ],
+                    bubble_full_width=True,
+                    type="messages"
+                )
+            with gr.Column():
+                gr.Markdown("### Prompt Optimized Chat")
+                chatbot_opt = gr.Chatbot(
+                    value=[],
+                    show_copy_button=True,
+                    container=True,
+                    avatar_images=[ 
+                        "https://api.holabo.co/user.svg", 
+                        "https://api.holabo.co/oxy.svg"
+                    ],
+                    bubble_full_width=True,
+                    type="messages"
+                )
+        # -------------------------
+        # Input + Sliders + Output
+        # -------------------------
         with gr.Row():
             msg = gr.Textbox(
                 label="Message",
                 placeholder="Type your message here...",
                 show_label=False,
-                container=False,
-                scale=9
+                container=True
             )
-            submit = gr.Button("Send", variant="primary", scale=1)
-
-        with gr.Accordion("Settings", open=False):
+            submit = gr.Button("Send", variant="primary")
+        with gr.Row():
             temperature = gr.Slider(
                 minimum=0.1,
                 maximum=2.0,
@@ -224,40 +292,45 @@ with gr.Blocks(
                 step=0.05,
                 label="Top-p"
             )
-        stats_display = gr.Markdown()
+        with gr.Row():
+            stats_display_base = gr.Markdown()
+            stats_display_opt = gr.Markdown()
 
-    def show_chat():
-        return gr.update(visible=False), gr.update(visible=True)
+        # -----------------------------
+        # Manage Chat Histories (State)
+        # -----------------------------
+        state_base = gr.State([])
+        state_opt = gr.State([])
 
-    msg.submit(
-        predict,
-        [msg, chatbot, temperature, top_p],
-       [chatbot, stats_display]
-    ).then(
-        lambda: "",
-        None,
-        msg
-    )
+        # -------------------------------------
+        # Update Both Chats On Submit
+        # -------------------------------------
+        msg.submit(
+            predict_synced,
+            [msg, state_base, state_opt, temperature, top_p],
+            [chatbot_base, stats_display_base, chatbot_opt, stats_display_opt]
+        ).then(lambda: "", None, msg)
+        submit.click(
+            predict_synced,
+            [msg, state_base, state_opt, temperature, top_p],
+            [chatbot_base, stats_display_base, chatbot_opt, stats_display_opt]
+        ).then(lambda: "", None, msg)
     
-    submit.click(
-        predict,
-        [msg, chatbot, temperature, top_p],
-        [chatbot, stats_display]
-    ).then(
-        lambda: "",
-        None,
-        msg
-    )
-
+    # ------------------
+    # Consent -> Chat UI
+    # ------------------
     agree_button.click(
         show_chat,
         inputs=None,
         outputs=[consent_block, chat_block]
     )
 
+# ----------------------
+# Launch the Gradio App
+# ----------------------
 if __name__ == "__main__":
     demo.launch(
         server_name="0.0.0.0",
         server_port=7860,
         share=False
-    )
+    )