import os import gradio as gr from openai import OpenAI from optillm.cot_reflection import cot_reflection from optillm.rto import round_trip_optimization from optillm.z3_solver import Z3SolverSystem from optillm.self_consistency import advanced_self_consistency_approach from optillm.rstar import RStar from optillm.plansearch import plansearch from optillm.leap import leap from optillm.reread import re2_approach API_KEY = os.environ.get("OPENROUTER_API_KEY") def compare_responses(message, model1, approach1, model2, approach2, system_message, max_tokens, temperature, top_p): response1 = respond(message, [], model1, approach1, system_message, max_tokens, temperature, top_p) response2 = respond(message, [], model2, approach2, system_message, max_tokens, temperature, top_p) return response1, response2 def parse_conversation(messages): system_prompt = "" conversation = [] for message in messages: role = message['role'] content = message['content'] if role == 'system': system_prompt = content elif role in ['user', 'assistant']: conversation.append(f"{role.capitalize()}: {content}") initial_query = "\n".join(conversation) return system_prompt, initial_query def respond(message, history, model, approach, system_message, max_tokens, temperature, top_p): client = OpenAI(api_key=API_KEY, base_url="https://openrouter.ai/api/v1") messages = [{"role": "system", "content": system_message}] for val in history: if val[0]: messages.append({"role": "user", "content": val[0]}) if val[1]: messages.append({"role": "assistant", "content": val[1]}) messages.append({"role": "user", "content": message}) if approach == "none": response = client.chat.completions.create( extra_headers={ "HTTP-Referer": "https://github.com/codelion/optillm", "X-Title": "optillm" }, model=model, messages=messages, max_tokens=max_tokens, temperature=temperature, top_p=top_p, ) return response.choices[0].message.content else: system_prompt, initial_query = parse_conversation(messages) if approach == 'rto': final_response, _ = round_trip_optimization(system_prompt, initial_query, client, model) elif approach == 'z3': z3_solver = Z3SolverSystem(system_prompt, client, model) final_response, _ = z3_solver.process_query(initial_query) elif approach == "self_consistency": final_response, _ = advanced_self_consistency_approach(system_prompt, initial_query, client, model) elif approach == "rstar": rstar = RStar(system_prompt, client, model) final_response, _ = rstar.solve(initial_query) elif approach == "cot_reflection": final_response, _ = cot_reflection(system_prompt, initial_query, client, model) elif approach == 'plansearch': final_response, _ = plansearch(system_prompt, initial_query, client, model)[0] elif approach == 'leap': final_response, _ = leap(system_prompt, initial_query, client, model) elif approach == 're2': final_response, _ = re2_approach(system_prompt, initial_query, client, model) return final_response # for message in client.chat_completion( # messages, # max_tokens=max_tokens, # stream=True, # temperature=temperature, # top_p=top_p, # ): # token = message.choices[0].delta.content # response += token # yield response def create_model_dropdown(): return gr.Dropdown( [ "meta-llama/llama-3.1-8b-instruct:free", "nousresearch/hermes-3-llama-3.1-405b:free", "mistralai/mistral-7b-instruct:free","mistralai/pixtral-12b:free", "qwen/qwen-2-7b-instruct:free", "qwen/qwen-2-vl-7b-instruct:free", "google/gemma-2-9b-it:free", "google/gemini-flash-8b-1.5-exp", "google/gemini-flash-1.5-exp", "google/gemini-pro-1.5-exp"], value="meta-llama/llama-3.1-8b-instruct:free", label="Model" ) def create_approach_dropdown(): return gr.Dropdown( ["none", "leap", "plansearch", "rstar", "cot_reflection", "rto", "self_consistency", "z3", "re2"], value="none", label="Approach" ) html = """ """ with gr.Blocks() as demo: gr.Markdown("# optillm - LLM Optimization Comparison") gr.HTML(html) with gr.Row(): system_message = gr.Textbox(value="", label="System message") max_tokens = gr.Slider(minimum=1, maximum=4096, value=1024, step=1, label="Max new tokens") temperature = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature") top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)") with gr.Tabs(): with gr.TabItem("Chat"): model = create_model_dropdown() approach = create_approach_dropdown() chatbot = gr.Chatbot() msg = gr.Textbox() with gr.Row(): submit = gr.Button("Submit") clear = gr.Button("Clear") def user(user_message, history): return "", history + [[user_message, None]] def bot(history, model, approach, system_message, max_tokens, temperature, top_p): user_message = history[-1][0] bot_message = respond(user_message, history[:-1], model, approach, system_message, max_tokens, temperature, top_p) history[-1][1] = bot_message return history msg.submit(user, [msg, chatbot], [msg, chatbot]).then( bot, [chatbot, model, approach, system_message, max_tokens, temperature, top_p], chatbot ) submit.click(user, [msg, chatbot], [msg, chatbot]).then( bot, [chatbot, model, approach, system_message, max_tokens, temperature, top_p], chatbot ) clear.click(lambda: None, None, chatbot, queue=False) with gr.TabItem("Compare"): with gr.Row(): model1 = create_model_dropdown() approach1 = create_approach_dropdown() model2 = create_model_dropdown() approach2 = create_approach_dropdown() compare_input = gr.Textbox(label="Enter your message for comparison") compare_button = gr.Button("Compare") with gr.Row(): output1 = gr.Textbox(label="Response 1") output2 = gr.Textbox(label="Response 2") compare_button.click( compare_responses, inputs=[compare_input, model1, approach1, model2, approach2, system_message, max_tokens, temperature, top_p], outputs=[output1, output2] ) if __name__ == "__main__": demo.launch()