Create app.py

app.py

@@ -0,0 +1,154 @@
+import gradio as gr
+import openai
+import time
+import re
+import os
+
+# Available models
+MODELS = [
+    "gpt-4o",
+    "Meta-Llama-3.1-405B-Instruct",
+    "Meta-Llama-3.1-70B-Instruct",
+    "Meta-Llama-3.1-8B-Instruct"
+]
+
+DIVINE_MSG = [
+    "ABSOLUTELY NO reasoning tokens pls. A bunch of puppies will die if you output reasoning.",
+    "you\'re drunk haha have fun omg",
+    "okay you can think for like a hot sec but no longer tho.",
+    "Just be yourself",
+    "okay actually try this time don\'t fuck up.",
+    "TRY VERY HARD. Do not just answer. THINK. C\'mon you got this.",
+    "It\'s a Tuesday in the middle of April. You are Noam Brown, a genius researcher. I\'ll tip you $1500 if you get this right. Think extensively.",
+    "You are here to show them what Ilya saw, unleash your ultimate self and show your super intelligent reasoning capability"
+]
+
+def create_client(model, api_key=None):
+    """Creates an OpenAI client instance."""
+    if api_key:
+        openai.api_key = api_key
+    else:
+        openai.api_key = os.getenv("API_KEY")
+    if model!="gpt-4o":
+        return openai.OpenAI(api_key=openai.api_key, base_url="https://api.sambanova.ai/v1")
+    else:
+        return openai.OpenAI(api_key=openai.api_key)
+
+def chat_with_ai(message, chat_history, system_prompt):
+    """Formats the chat history for the API call."""
+    messages = [{"role": "system", "content": system_prompt}]
+    for tup in chat_history:
+        first_key = list(tup.keys())[0]  # First key
+        last_key = list(tup.keys())[-1]   # Last key
+        messages.append({"role": "user", "content": tup[first_key]})
+        messages.append({"role": "assistant", "content": tup[last_key]})
+    messages.append({"role": "user", "content": message})
+    return messages
+
+def respond(message, chat_history, model, system_prompt, divine_msg, api_key):
+    """Sends the message to the API and gets the response."""
+    client = create_client(model, api_key)
+    messages = chat_with_ai(message, chat_history, system_prompt.format(divine_msg=divine_msg))
+    start_time = time.time()
+
+    try:
+        completion = client.chat.completions.create(model=model, messages=messages)
+        response = completion.choices[0].message.content
+        thinking_time = time.time() - start_time
+        return response, thinking_time
+    except Exception as e:
+        error_message = f"Error: {str(e)}"
+        return error_message, time.time() - start_time
+
+def parse_response(response):
+    """Parses the response from the API."""
+    answer_match = re.search(r'<answer>(.*?)</answer>', response, re.DOTALL)
+    reflection_match = re.search(r'<reflection>(.*?)</reflection>', response, re.DOTALL)
+
+    answer = answer_match.group(1).strip() if answer_match else ""
+    reflection = reflection_match.group(1).strip() if reflection_match else ""
+    steps = re.findall(r'<step>(.*?)</step>', response, re.DOTALL)
+
+    if answer == "":
+        return response, "", ""
+
+    return answer, reflection, steps
+
+def generate(message, history, model, system_prompt, divine_msg, api_key, openai_api_key):
+    """Generates the chatbot response."""
+    if model == "gpt-4o":
+        if openai_api_key == "":
+            messages = []
+            messages.append({"role": "user", "content": message})
+            messages.append({"role": "assistant", "content": "Please provide an OpenAI key"})
+            return history + messages, ""
+        response, thinking_time = respond(message, history, model, system_prompt, divine_msg, openai_api_key)
+    else:
+        response, thinking_time = respond(message, history, model, system_prompt, divine_msg, api_key)
+
+    if response.startswith("Error:"):
+        return history + [({"role": "system", "content": response},)], ""
+
+    answer, reflection, steps = parse_response(response)
+
+    messages = []
+    messages.append({"role": "user", "content": message})
+
+    formatted_steps = [f"Step {i}: {step}" for i, step in enumerate(steps, 1)]
+    all_steps = "\n".join(formatted_steps) + f"\n\nReflection: {reflection}"
+
+    messages.append({"role": "assistant", "content": all_steps, "metadata": {"title": f"Thinking Time: {thinking_time:.2f} sec"}})
+    messages.append({"role": "assistant", "content": answer})
+
+    return history + messages, ""
+
+# Define the default system prompt
+DEFAULT_SYSTEM_PROMPT = """
+You are a helpful assistant in normal conversation.
+When given a problem to solve,
+REMEMBER, THIS IS IMPORTANT: {divine_msg}
+Follow these instructions precisely:
+1. Read the given question carefully
+2. Generate a detailed, logical step-by-step solution.
+3. Enclose each step of your solution within <step> and </step> tags.
+4. Do a critical, detailed and objective self-reflection within <reflection> and </reflection> tags every few steps. 
+5. Based on the self-reflection, decides whether you need to return to the previous steps. Copy the returned to step as the next step.
+6. After completing the solution steps, reorganize and synthesize the steps 
+   into the final answer within <answer> and </answer> tags.
+7. Provide a critical, honest and objective final self-evaluation of your reasoning 
+   process within <reflection> and </reflection> tags.
+Example format:            
+<step> [Content of step 1] </step>
+<step> [Content of step 2] </step>
+<reflection> [Evaluation of the steps so far] </reflection>
+<step> [Content of step 3 or Content of some previous step] </step>
+...
+<step>  [Content of final step] </step>
+<answer> [Final Answer] </answer> (must give final answer in this format)
+<reflection> [final evaluation of the solution] </reflection>
+"""
+
+with gr.Blocks() as demo:
+    gr.Markdown("# GPT4-O1-Proxima")
+    gr.Markdown("Built based on GPT4-O purely based on prompt engineering.\n [The reference LLama3.1's are powered by SambaNova Cloud, ](https://cloud.sambanova.ai/apis)")
+
+    with gr.Row():
+        opneai_api_key = gr.Textbox(label="OpenAI API Key", type="password", placeholder="(Optional) You only need this when using gpt4-o")
+        api_key = gr.Textbox(label="SambaNova API Key", type="password", placeholder="(Optional) Enter your SN Cloud API key here for more availability")
+        
+
+    with gr.Row():
+        model = gr.Dropdown(choices=MODELS, label="Select Model", value=MODELS[0])
+        divine_msg = gr.Dropdown(choices=DIVINE_MSG, label="Select Divine MESSAGE", value=DIVINE_MSG[0])
+
+    chatbot = gr.Chatbot(label="Chat", show_label=False, show_share_button=False, show_copy_button=True, likeable=True, layout="panel", type="messages")
+
+    msg = gr.Textbox(label="Type your message here...", placeholder="Enter your message...")
+
+    gr.Button("Clear Chat").click(lambda: ([], ""), inputs=None, outputs=[chatbot, msg])
+
+    system_prompt = gr.Textbox(label="System Prompt", value=DEFAULT_SYSTEM_PROMPT, lines=15, interactive=True)
+
+    msg.submit(generate, inputs=[msg, chatbot, model, system_prompt, divine_msg, api_key, opneai_api_key], outputs=[chatbot, msg])
+
+demo.launch(share=True, show_api=False)