app.py

import gradio
import argparse
import os
import boto3
import pandas as pd
from copy import copy

import queue

from constants import css, js_code, js_light
from utils import model_response, clear_chat
from models import get_tinyllama, get_qwen2ins1b, GigaChat, response_gigachat, response_qwen2ins1b, response_tinyllama

INIT_MODELS = dict()
S3_SESSION = None
TEST_MD = None
CURRENT_MODELS = queue.LifoQueue()
MODEL_LIB = {'TINYLLAMA': get_tinyllama, "QWEN2INS1B": get_qwen2ins1b, "RUBASE": GigaChat.get_giga}
GEN_LIB = {'TINYLLAMA': response_tinyllama, "QWEN2INS1B": response_qwen2ins1b, "RUBASE": response_gigachat}

def model_gen(
        content,
        chat_history,
        model_name: str,
        top_p,
        temp,
        max_tokens,
        no_context=False
        ):
    
    global INIT_MODELS, S3_SESSION, GEN_LIB, MODEL_LIB
    model_manager(model_name, MODEL_LIB, 3)
    if content is None:
        return '', []
    if len(content) == 0:
        return '', []
    
    chat_history = [chat_history[-1]] if no_context and len(chat_history)>=1  else chat_history
    
    return model_response(
        content, 
        chat_history,
        S3_SESSION, 
        INIT_MODELS,
        GEN_LIB,
        model_name, 
        {"top_p": top_p, "temperature": temp, "max_tokens": max_tokens}
        )

def model_regen(
        content,
        chat_history,
        model_name: str,
        top_p,
        temp,
        max_tokens,
        no_context=False
        ):
    
    global INIT_MODELS, S3_SESSION, GEN_LIB, MODEL_LIB
    model_manager(model_name, MODEL_LIB, 3)
    if chat_history is None:
        return '', []
    
    chat_history = chat_history[-1] if no_context else chat_history
    content = copy(chat_history[-1][0])
    
    return model_response(
        content, 
        chat_history[:-1],
        S3_SESSION, 
        INIT_MODELS,
        GEN_LIB,
        model_name, 
        {"top_p": top_p, "temperature": temp, "max_tokens": max_tokens}
        )

def model_manager(
        add_model,
        model_lib,
        max_models=3
        ):
    global INIT_MODELS, CURRENT_MODELS
    while CURRENT_MODELS.qsize() >= max_models:
        model_del = CURRENT_MODELS.get()
        INIT_MODELS[model_del] = None
    CURRENT_MODELS.put(add_model)
    INIT_MODELS[add_model] = model_lib[add_model]()

def tab_online_arena():
    global S3_SESSION, GEN_LIB, MODEL_LIB, INIT_MODELS, CURRENT_MODELS
    with gradio.Row():
        with gradio.Column():
            model_left = gradio.Dropdown(["TINYLLAMA", "QWEN2INS1B", "RUBASE"], value="QWEN2INS1B", interactive=True, multiselect=False, label="Left model")
            chatbot_left = gradio.Chatbot()
        with gradio.Column():
            model_right = gradio.Dropdown(["TINYLLAMA", "QWEN2INS1B", "RUBASE"], value= "RUBASE", interactive=True, multiselect=False, label="Right model")
            chatbot_right = gradio.Chatbot()

    with gradio.Row():
        msg = gradio.Textbox(label='Prompt', placeholder='Put your prompt here')

    with gradio.Row():
        gradio.Button('Both Good')
        gradio.Button('Left Better')
        gradio.Button('Right Better')
        gradio.Button('Both Bad')

    with gradio.Row():
        with gradio.Accordion("Parameters", open=False):
            no_context = gradio.Checkbox(label="No context", value=False)
            top_p = gradio.Slider(label='Top P', minimum=0, maximum=1, value=1, step=0.05, interactive=True)
            temp = gradio.Slider(label='Temperature', minimum=0, maximum=1, value=0.7, step=0.05, interactive=True)
            max_tokens = gradio.Slider(label='Max ouput tokens', minimum=1, maximum=2048, value=256, step=1, interactive=True)

    with gradio.Row():
        clear = gradio.ClearButton([msg, chatbot_left, chatbot_right], value='Clear history')
        regen_left = gradio.Button(value='Regenerate left answer')
        regen_right = gradio.Button(value='Regenerate right answer')
        regen_left.click(
            model_regen, 
            [msg, chatbot_left, model_left, top_p, temp, max_tokens, no_context], 
             [msg, chatbot_left]
             )
        regen_right.click(
            model_regen,
            [msg, chatbot_right, model_right, top_p, temp, max_tokens, no_context], 
             [msg, chatbot_right]
             )
    
    with gradio.Blocks():
        model_left.change(clear_chat, [], [msg, chatbot_left])
        model_right.change(clear_chat, [], [msg, chatbot_right])
        msg.submit(
            model_gen, 
            [msg, chatbot_left, model_left, top_p, temp, max_tokens, no_context], 
             [msg, chatbot_left]
             )
        msg.submit(
            model_gen,
            [msg, chatbot_right, model_right, top_p, temp, max_tokens, no_context], 
             [msg, chatbot_right]
             )

        # with gradio.Column():
        #     gradio.ChatInterface(
        #         fn=giga_gen,
        #         examples=[{"text": "hello"}, {"text": "hola"}, {"text": "merhaba"}],
        #         title="Giga",
        #         multimodal=True,
        #     )
        # with gradio.Column():
        #     gradio.ChatInterface(
        #         fn=tiny_gen,
        #         examples=[{"text": "hello"}, {"text": "hola"}, {"text": "merhaba"}],
        #         title="Tiny",
        #         multimodal=True,
        #     )
        # with gradio.Column():
        #     gradio.Interface(fn=giga_gen, inputs="text", outputs="text", allow_flagging=False, title='Giga') # arena = 
        # with gradio.Column():
        #     gradio.Interface(fn=tiny_gen, inputs="text", outputs="text", allow_flagging=False, title='TinyLlama') # arena = 
        
    # arena.launch()

def tab_leaderboard():
    df = pd.DataFrame({
            "Model" : ['A', 'B', 'C',], 
            "Test 1" : [0, 1, 0], 
            "Test 2" : [1, 0, 1,],
        })

    # Function to apply text color
    def highlight_cols(x): 
        df = x.copy() 
        # df.loc[:, :] = 'color: purple'
        df[['Model']] = 'color: green'
        return df 

    # Applying the style function
    # s = df.style.apply(highlight_cols, axis = None)

    # Displaying the styled dataframe in Gradio
    with gradio.TabItem("Autogen Metrics", elem_id="od-benchmark-tab-table-ablation", id=0, elem_classes="subtab"): 
        with gradio.Blocks() as demo:
            gradio.DataFrame(df)
    
    with gradio.TabItem("Autometrics", elem_id="od-benchmark-tab-table-ablation", id=1, elem_classes="subtab"): 
        with gradio.Blocks() as demo:
            gradio.DataFrame(df)
    
    with gradio.TabItem("SBS metrics", elem_id="od-benchmark-tab-table-ablation", id=2, elem_classes="subtab"): 
        with gradio.Blocks() as demo:
            gradio.DataFrame(df)
    
    with gradio.TabItem("Arena ELO rating", elem_id="od-benchmark-tab-table-ablation", id=3, elem_classes="subtab"): 
        with gradio.Blocks() as demo:
            gradio.DataFrame(df)

def tab_offline_arena(): 
    # with gradio.Row(): 
    #     btn_show_history = gradio.Button("🎲  Click here to sample an example + a pair of LLM outputs! ", elem_classes="sample_button")
    with gradio.Row():
        with gradio.Column(scale=2):
            with gradio.Accordion("Choose models to sample from", open=False, elem_classes="accordion-label"):
                model_options = list(MODEL_LIB.keys())  
                selected_models = gradio.CheckboxGroup(model_options, info="", value=model_options, show_label=False, elem_id="select-models") 
                clear_button = gradio.Button("Clear", elem_classes="btn_boderline_gray", scale=1)
                # clear the selected_models
                clear_button.click(lambda: {selected_models: {"value": [], "__type__": "update"}}, inputs=[], outputs=[selected_models])
        with gradio.Column(scale=1):
            with gradio.Accordion("Choose task types to sample from", open=False, elem_classes="accordion-label"):
                select_tasks = gradio.CheckboxGroup(['Task 1', "Task 2", "Task 3"], info="", value=['Task 1', "Task 2", "Task 3"], show_label=False, elem_id="select-tasks") 
                clear_task_button = gradio.Button("Clear", elem_classes="btn_boderline_gray", scale=1)
                # clear the select_tasks
                clear_task_button.click(lambda: {select_tasks: {"value": [], "__type__": "update"}}, inputs=[], outputs=[select_tasks])
            with gradio.Accordion("Choose criteria  to sample from", open=False, elem_classes="accordion-label"):
                select_tasks = gradio.CheckboxGroup(['Criterion 1', "Criterion 2", "Criterion 3"], info="", value=['Criterion 1', "Criterion 2", "Criterion 3"], show_label=False, elem_id="select-criteria") 
                clear_task_button = gradio.Button("Clear", elem_classes="btn_boderline_gray", scale=1)
                # clear the select_tasks
                clear_task_button.click(lambda: {select_tasks: {"value": [], "__type__": "update"}}, inputs=[], outputs=[select_tasks])


def build_demo():
    # global original_dfs, available_models, gpt4t_dfs, haiku_dfs, llama_dfs

    with gradio.Blocks(theme=gradio.themes.Base(), css=css, js=js_light) as demo:
        # gradio.HTML(BANNER, elem_id="banner")
        # gradio.Markdown(HEADER_MD.replace("{model_num}", str(len(original_dfs["-1"]))), elem_classes="markdown-text")
        
        with gradio.Tabs(elem_classes="tab-buttons") as tabs:
            with gradio.TabItem("🐼 MERA leaderboard", elem_id="od-benchmark-tab-table", id=0):
                gradio.Markdown(TEST_MD, elem_classes="markdown-text-details")
                tab_leaderboard()

            with gradio.TabItem("🆚 SBS by categories and criteria", elem_id="od-benchmark-tab-table", id=1):
                # gradio.Markdown(TEST_MD, elem_classes="markdown-text-details")
                tab_offline_arena()

            with gradio.TabItem("🥊 Model arena", elem_id="od-benchmark-tab-table", id=2):
                tab_online_arena()
                # _tab_explore()

            with gradio.TabItem("💪 About MERA", elem_id="od-benchmark-tab-table", id=3):
                gradio.Markdown(TEST_MD, elem_classes="markdown-text")
        # gr.Markdown(f"Last updated on **{LAST_UPDATED}** | [Link to V1-legacy](https://huggingface.co/spaces/allenai/WildBench-V1-legacy)", elem_classes="markdown-text-small")
        
        # with gr.Row():
        #     with gr.Accordion("📙 Citation", open=False, elem_classes="accordion-label"):
        #         gr.Textbox(
        #             value=CITATION_TEXT, 
        #             lines=7,
        #             label="Copy the BibTeX snippet to cite this source",
        #             elem_id="citation-button",
        #             show_copy_button=True)
                # ).style(show_copy_button=True)

    return demo

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--share", action="store_true")
    # parser.add_argument("--bench_table", help="Path to MERA table", default="data_dir/MERA_jun2024.jsonl")
    args = parser.parse_args()
    # data_load(args.result_file)    
    # TYPES = ["number", "markdown", "number"]

    with open("test.md", "r") as f:
        TEST_MD = f.read()

    try:
        session = boto3.session.Session()
        S3_SESSION = session.client(
            service_name='s3',
            endpoint_url=os.getenv('S3_ENDPOINT'),
            aws_access_key_id=os.getenv('S3_ACCESS_KEY'),
            aws_secret_access_key=os.getenv('S3_SECRET_KEY'),
        )
    except:
        print('Failed to start s3 session')

    app = build_demo()
    app.launch(share=args.share, height=3000, width="110%") # share=args.share

    # demo = gradio.Interface(fn=gen, inputs="text", outputs="text")
    # demo.launch()