app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM
import time
import os
import openai

# Load the Vicuna 7B v1.3 LMSys model and tokenizer
model_name = "lmsys/vicuna-7b-v1.3"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

template_single = '''Please output any <{}> in the following sentence one per line without any additional text: "{}"'''

def update_api_key(new_key):
    global api_key
    os.environ['OPENAI_API_TOKEN'] = new_key
    openai.api_key = os.environ['OPENAI_API_TOKEN']

def chat(system_prompt, user_prompt, model = 'gpt-3.5-turbo', temperature = 0, verbose = False):
    ''' Normal call of OpenAI API '''
    response = openai.ChatCompletion.create(
    temperature = temperature,
    model=model,
    messages=[
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": user_prompt}
    ])
    
    res = response['choices'][0]['message']['content']
    
    if verbose:
        print('System prompt:', system_prompt)
        print('User prompt:', user_prompt)
        print('GPT response:', res)
        
    return res

def format_chat_prompt(message, chat_history, max_convo_length):
    prompt = ""
    for turn in chat_history[-max_convo_length:]:
        user_message, bot_message = turn
        prompt = f"{prompt}\nUser: {user_message}\nAssistant: {bot_message}"
    prompt = f"{prompt}\nUser: {message}\nAssistant:"
    return prompt

def respond_gpt(tab_name, message, chat_history, max_convo_length = 10):
        formatted_prompt = format_chat_prompt(message, chat_history, max_convo_length)
        print('Prompt + Context:')
        print(formatted_prompt)
        bot_message = chat(system_prompt = f'''Generate the output only for the assistant. Please output any <{tab_name}> in the following sentence one per line without any additional text.''',
                           user_prompt = formatted_prompt)
        chat_history.append((message, bot_message))
        return "", chat_history

def respond(message, chat_history):
    input_ids = tokenizer.encode(message, return_tensors="pt")
    output_ids = model.generate(input_ids, max_length=50, num_beams=5, no_repeat_ngram_size=2)
    bot_message = tokenizer.decode(output_ids[0], skip_special_tokens=True)
        
    chat_history.append((message, bot_message))
    time.sleep(2)
    return "", chat_history

def interface():
        gr.Markdown(" Description ")

        textbox_prompt = gr.Textbox(show_label=False, placeholder="Write a prompt and press enter")
        with gr.Row():
            api_key_input = gr.Textbox(label="Open AI Key", placeholder="Enter your Openai key here", type="password")
            api_key_btn = gr.Button(label="Submit Api Key", scale=0)
        tab_name = gr.Dropdown(["Noun", "Determiner", "Noun phrase", "Verb phrase", "Dependent clause", "T-units"], label="Linguistic Entity")
        btn = gr.Button(label="Submit")

        # prompt = template_single.format(tab_name, textbox_prompt)

        gr.Markdown("Strategy 1 QA-Based Prompting")
        with gr.Row():
            vicuna_S1_chatbot = gr.Chatbot(label="vicuna-7b")
            llama_S1_chatbot = gr.Chatbot(label="llama-7b")
            gpt_S1_chatbot = gr.Chatbot(label="gpt-3.5")
        clear = gr.ClearButton(components=[textbox_prompt, vicuna_S1_chatbot])
        gr.Markdown("Strategy 2 Instruction-Based Prompting")
        with gr.Row():
            vicuna_S2_chatbot = gr.Chatbot(label="vicuna-7b")
            llama_S2_chatbot = gr.Chatbot(label="llama-7b")
            gpt_S2_chatbot = gr.Chatbot(label="gpt-3.5")
        clear = gr.ClearButton(components=[textbox_prompt, vicuna_S2_chatbot])
        gr.Markdown("Strategy 3 Structured Prompting")
        with gr.Row():
            vicuna_S3_chatbot = gr.Chatbot(label="vicuna-7b")
            llama_S3_chatbot = gr.Chatbot(label="llama-7b")
            gpt_S3_chatbot = gr.Chatbot(label="gpt-3.5")
        clear = gr.ClearButton(components=[textbox_prompt, vicuna_S3_chatbot])

        textbox_prompt.submit(respond, inputs=[textbox_prompt, vicuna_S1_chatbot], outputs=[textbox_prompt, vicuna_S1_chatbot])
        textbox_prompt.submit(respond, inputs=[textbox_prompt, vicuna_S2_chatbot], outputs=[textbox_prompt, vicuna_S2_chatbot])
        textbox_prompt.submit(respond, inputs=[textbox_prompt, vicuna_S3_chatbot], outputs=[textbox_prompt, vicuna_S3_chatbot])

        api_key_btn.click(update_api_key, inputs=api_key_input)
        btn.click(respond_gpt, inputs=[tab_name, textbox_prompt, gpt_S1_chatbot], outputs=[tab_name, textbox_prompt, gpt_S1_chatbot])

with gr.Blocks() as demo:
    gr.Markdown("# LLM Evaluator With Linguistic Scrutiny")

    with gr.Tab("Noun"):
        interface()
    
    with gr.Tab("Determiner"):
        gr.Markdown(" Description ")

        prompt_CHUNK = gr.Textbox(show_label=False, placeholder="Write a prompt and press enter")

        gr.Markdown("Strategy 1 QA")
        with gr.Row():
            vicuna_S1_chatbot_CHUNK = gr.Chatbot(label="vicuna-7b")
            llama_S1_chatbot_CHUNK = gr.Chatbot(label="llama-7b")
            gpt_S1_chatbot_CHUNK = gr.Chatbot(label="gpt-3.5")
        clear = gr.ClearButton([prompt_CHUNK, vicuna_S1_chatbot_CHUNK])
        gr.Markdown("Strategy 2 Instruction")
        with gr.Row():
            vicuna_S2_chatbot_CHUNK = gr.Chatbot(label="vicuna-7b")
            llama_S2_chatbot_CHUNK = gr.Chatbot(label="llama-7b")
            gpt_S2_chatbot_CHUNK = gr.Chatbot(label="gpt-3.5")
        clear = gr.ClearButton([prompt_CHUNK, vicuna_S2_chatbot_CHUNK])
        gr.Markdown("Strategy 3 Structured Prompting")
        with gr.Row():
            vicuna_S3_chatbot_CHUNK = gr.Chatbot(label="vicuna-7b")
            llama_S3_chatbot_CHUNK = gr.Chatbot(label="llama-7b")
            gpt_S3_chatbot_CHUNK = gr.Chatbot(label="gpt-3.5")
        clear = gr.ClearButton([prompt_CHUNK, vicuna_S3_chatbot_CHUNK])
    
    with gr.Tab("Noun phrase"):
        interface()
    with gr.Tab("Verb phrase"):
        interface()
    with gr.Tab("Dependent clause"):
        interface()
    with gr.Tab("T-units"):
        interface()

    prompt_CHUNK.submit(respond, [prompt_CHUNK, vicuna_S1_chatbot_CHUNK], [prompt_CHUNK, vicuna_S1_chatbot_CHUNK])
    prompt_CHUNK.submit(respond, [prompt_CHUNK, vicuna_S2_chatbot_CHUNK], [prompt_CHUNK, vicuna_S2_chatbot_CHUNK])
    prompt_CHUNK.submit(respond, [prompt_CHUNK, vicuna_S3_chatbot_CHUNK], [prompt_CHUNK, vicuna_S3_chatbot_CHUNK])

demo.launch()