app.py

import gradio as gr
import time 
import json
from cerebras.cloud.sdk import Cerebras
from typing import List, Dict, Tuple, Any, Generator
from tenacity import retry, stop_after_attempt, wait_fixed

def make_api_call(api_key: str, messages: List[Dict[str, str]], max_tokens: int, is_final_answer: bool = False) -> Dict[str, Any]:
    """
    Make an API call to the Cerebras chat completions endpoint with retry logic.
    """
    client = Cerebras(api_key=api_key)
    
    try:
        start_time = time.time()
        response = client.chat.completions.create(
            model="llama3.1-70b",
            messages=messages,
            max_tokens=max_tokens,
            temperature=0.2,
            response_format={"type": "json_object"}
        )
        end_time = time.time()
        
        content = json.loads(response.choices[0].message.content)
        
        # Calculate tokens per second
        total_tokens = response.usage.total_tokens
        elapsed_time = end_time - start_time
        tokens_per_second = total_tokens / elapsed_time if elapsed_time > 0 else 0
        
        content['token_info'] = {
            'total_tokens': total_tokens,
            'tokens_per_second': tokens_per_second
        }
        
        return content
    except Exception as e:
        if is_final_answer:
            return {"title": "Error", "content": f"Failed to generate final answer. Error: {str(e)}"}
        else:
            return {"title": "Error", "content": f"Failed to generate step. Error: {str(e)}", "next_action": "final_answer"}

def generate_response(api_key: str, prompt: str) -> Generator[Tuple[List[Tuple[str, str]], float, int, float], None, None]:
    """
    Generate a response to the given prompt using a step-by-step reasoning approach.
    This function is now a generator that yields each step as it's generated.
    """
    system_message = """You are an expert AI assistant that explains your reasoning step by step. For each step, provide a title that describes what you're doing in that step, along with the content. Decide if you need another step or if you're ready to give the final answer. Respond in JSON format with 'title', 'content', and 'next_action' (either 'continue' or 'final_answer') keys. USE AS MANY REASONING STEPS AS POSSIBLE. AT LEAST 3. BE AWARE OF YOUR LIMITATIONS AS AN LLM AND WHAT YOU CAN AND CANNOT DO. IN YOUR REASONING, INCLUDE EXPLORATION OF ALTERNATIVE ANSWERS. CONSIDER YOU MAY BE WRONG, AND IF YOU ARE WRONG IN YOUR REASONING, WHERE IT WOULD BE. FULLY TEST ALL OTHER POSSIBILITIES. YOU CAN BE WRONG. WHEN YOU SAY YOU ARE RE-EXAMINING, ACTUALLY RE-EXAMINE, AND USE ANOTHER APPROACH TO DO SO. DO NOT JUST SAY YOU ARE RE-EXAMINING. USE AT LEAST 3 METHODS TO DERIVE THE ANSWER. USE BEST PRACTICES."""
    
    messages = [
        {"role": "system", "content": system_message},
        {"role": "user", "content": prompt},
        {"role": "assistant", "content": "Thank you! I will now think step by step following my instructions, starting at the beginning after decomposing the problem."}
    ]
    
    steps = []
    step_count = 1
    total_thinking_time = 0
    total_tokens = 0
    total_tokens_per_second = 0
    
    while True:
        start_time = time.time()
        step_data = make_api_call(api_key, messages, 300)
        thinking_time = time.time() - start_time
        total_thinking_time += thinking_time
        
        token_info = step_data.pop('token_info', {'total_tokens': 0, 'tokens_per_second': 0})
        total_tokens += token_info['total_tokens']
        total_tokens_per_second += token_info['tokens_per_second']
        
        step_title = f"Step {step_count}: {step_data['title']}"
        step_content = f"{step_data['content']}\n\n**Cerebras LLM Call Duration: {thinking_time:.2f} seconds**\n**Tokens: {token_info['total_tokens']}, Tokens/s: {token_info['tokens_per_second']:.2f}**"
        steps.append((step_title, step_content))
        messages.append({"role": "assistant", "content": json.dumps(step_data)})
        
        # Yield the current conversation, total thinking time, total tokens, and average tokens per second
        yield steps, total_thinking_time, total_tokens, total_tokens_per_second / step_count if step_count > 0 else 0
        
        if step_data.get('next_action') == 'final_answer':
            break
        
        step_count += 1

    # Request the final answer
    messages.append({"role": "user", "content": "Please provide the final answer based on your reasoning above."})
    
    start_time = time.time()
    final_data = make_api_call(api_key, messages, 200, is_final_answer=True)
    thinking_time = time.time() - start_time
    total_thinking_time += thinking_time
    
    token_info = final_data.pop('token_info', {'total_tokens': 0, 'tokens_per_second': 0})
    total_tokens += token_info['total_tokens']
    total_tokens_per_second += token_info['tokens_per_second']
    
    final_content = f"{final_data.get('content', 'No final answer provided.')}\n\n**Final answer thinking time: {thinking_time:.2f} seconds**\n**Tokens: {token_info['total_tokens']}, Tokens/s: {token_info['tokens_per_second']:.2f}**"
    steps.append(("Final Answer", final_content))
    
    # Yield the final conversation, total thinking time, total tokens, and average tokens per second
    yield steps, total_thinking_time, total_tokens, total_tokens_per_second / (step_count + 1)

def respond(api_key: str, message: str, history: List[Tuple[str, str]]) -> Generator[Tuple[List[Tuple[str, str]], str], None, None]:
    """
    Generator function to handle responses and yield updates for streaming.
    """
    if not api_key:
        yield history, "Please provide a valid Cerebras API key."
        return

    # Initialize the generator
    response_generator = generate_response(api_key, message)
    
    for steps, total_time, total_tokens, avg_tokens_per_second in response_generator:
        conversation = history.copy()
        for title, content in steps[len(conversation):]:
            if title.startswith("Step") or title == "Final Answer":
                conversation.append((title, content))
            else:
                conversation.append((title, content))
        yield conversation, f"**Total thinking time:** {total_time:.2f} seconds\n**Total tokens:** {total_tokens}\n**Average tokens/s:** {avg_tokens_per_second:.2f}"

def main():
    with gr.Blocks() as demo:
        gr.Markdown("# o1-like Chain of Thought - LLaMA-3.1 70B on Cerebras")
        gr.Markdown("""
        Implement Chain of Thought with prompting to improve output accuracy.
        Powered by Cerebras, ensuring fast reasoning steps.
        """)
        
        with gr.Row():
            api_key_input = gr.Textbox(
                label="Cerebras API Key", 
                type="password", 
                placeholder="Enter your Cerebras API key", 
                show_label=True
            )
        
        chatbot = gr.Chatbot(label="Conversation")
        with gr.Row():
            user_input = gr.Textbox(
                label="Your Query", 
                placeholder="Enter your query here...", 
                show_label=True
            )
            submit_btn = gr.Button("Submit")
        
        thinking_time_display = gr.Textbox(
            label="Performance Metrics", 
            value="", 
            interactive=False
        )
        
        submit_btn.click(
            fn=respond, 
            inputs=[api_key_input, user_input, chatbot], 
            outputs=[chatbot, thinking_time_display],
            queue=True
        )
        
        # Allow pressing Enter to submit
        user_input.submit(
            fn=respond, 
            inputs=[api_key_input, user_input, chatbot], 
            outputs=[chatbot, thinking_time_display],
            queue=True
        )
    
    demo.launch()

if __name__ == "__main__":
    main()