app.py

"""A gradio app that renders a static leaderboard. This is used for Hugging Face Space."""
import ast
import argparse
import glob
import pickle
import plotly
import gradio as gr
import numpy as np
import pandas as pd
import gradio as gr
import pandas as pd
from pathlib import Path
import json
from constants import BANNER, CITATION_TEXT, WINRATE_HEATMAP, css, js_code, all_task_types, DEFAULT_LP, TASK_TYPE_STR, js_light
from datetime import datetime, timezone
from data_utils import load_eval_results, sample_an_eval_result, apply_length_penalty, post_processing, add_winrates, add_winrates_tasks
# from gradio.themes.utils import colors, fonts, sizes
from themes import Seafoam
from huggingface_hub import HfApi
# from datasets import Dataset, load_dataset, concatenate_datasets
import os, uuid 
from utils_display import model_info

# get the last updated time from the elo_ranks.all.jsonl file
LAST_UPDATED = None 
with open("_intro.md", "r") as f:
    INTRO_MD = f.read()

with open("_about_us.md", "r") as f:
    ABOUT_MD = f.read()

with open("_header.md", "r") as f:
    HEADER_MD = f.read()

original_df, ablation_df = None, None
eval_results = load_eval_results() 
 
available_models = [] # to be filled in later

  
def display_chat_history(model_selections):
    eval_item = sample_an_eval_result(eval_results, model_selections)
    task = eval_item["task"]
    if eval_item["image"]:
        image_path = eval_item["image"]
    else:
        image_path = ""
    chats_plan = []
    for item_user, item_asst in zip(eval_item["plan_history"]["user"], eval_item["plan_history"]["assistant"]):
        chats_plan += [item_user, item_asst]
    chats_ground = []
    for item_user, item_asst in zip(eval_item["ground_history"]["user"], eval_item["ground_history"]["assistant"]):
        chats_ground += [item_user, item_asst]
    chats_plan = [(chats_plan[i], chats_plan[i+1]) for i in range(0, len(chats_plan), 2)]
    chats_ground = [(chats_ground[i], chats_ground[i+1]) for i in range(0, len(chats_ground), 2)]
    
    if image_path != "":
        image = f'<div style="text-align: center;"> <img src="{image_path}" style="height: 250px;"> </div>'
        return task, chats_plan, chats_ground, image
    else:
        return task, chats_plan, chats_ground, f'<div style="text-align: center;"> </div>'




def slider_change_main(length_penalty):
    global original_df, ablation_df
    adjusted_df = apply_length_penalty(original_df, ablation_df, length_penalty) 
    adjusted_df = adjusted_df[["Model", "Overall Elo", "Task-Avg Elo", "# battles", "Length"]]
    adjusted_df = adjusted_df.sort_values(by="Overall Elo", ascending=False)
    adjusted_df = add_winrates(adjusted_df) 
    adjusted_df = adjusted_df.drop(columns=["Length"])
    return adjusted_df

def slider_change_full(length_penalty, show_winrate):
    global original_df, ablation_df
    adjusted_df = apply_length_penalty(original_df, ablation_df, length_penalty)
    # sort the model by the "Task-Avg Elo" column
    adjusted_df = adjusted_df.sort_values(by="Task-Avg Elo", ascending=False)
    adjusted_df.drop(columns=["Overall Elo", "Task-Avg Elo", "# battles", "Length"], inplace=True)
    if show_winrate == "none":
        return adjusted_df
    elif show_winrate == "gpt-3.5":
        adjusted_df = add_winrates_tasks(adjusted_df, ref="gpt-3.5")
    elif show_winrate == "gpt-4":
        adjusted_df = add_winrates_tasks(adjusted_df, ref="gpt-4")
    return adjusted_df

seafoam = Seafoam()
def build_demo(TYPES):
    global original_df, ablation_df, skip_empty_original_df, skip_empty_ablation_df, available_models
    with gr.Blocks(theme=gr.themes.Soft(), css=css, js=js_light) as demo:
        gr.Markdown(HEADER_MD, elem_classes="markdown-text")

        with gr.Tabs(elem_classes="tab-buttons") as tabs:
            with gr.TabItem("🔍 Explore", elem_id="od-benchmark-tab-table", id=2): 

                with gr.Row(): 
                    btn_show_history = gr.Button("🎲  Click here to sample an example of 🪄 Lumos outputs! ", elem_classes="sample_button")

                with gr.Row():
                    with gr.Column():
                        with gr.Accordion("Choose models to sample from", open=False, elem_classes="accordion-label"):
                            model_options = available_models
                            selected_models = gr.CheckboxGroup(model_options, info="", value=model_options, show_label=False, elem_id="select-models") 
                            clear_button = gr.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 gr.Row():
                    with gr.Column(scale=1.5):
                        with gr.Accordion("📝 Task Description", open=True, elem_classes="accordion-label"):
                            task = gr.Markdown("", elem_classes="markdown-text-tiny")
                            task.change(lambda x: x, inputs=[], outputs=[], scroll_to_output=False, js=js_code)
                    
                    with gr.Column(scale=1):
                        with gr.Accordion("Input Image (optional)", open=True, elem_classes="accordion-label"):
                            image = gr.HTML("", elem_id="input_image")
                            image.change(lambda x: x, inputs=[], outputs=[], scroll_to_output=False, js=js_code)

                with gr.Row():
                    with gr.Column():
                        gr.Markdown("## 📢 Plan Module Process History", elem_classes="markdown-text")
                        Chatbot_Common_Plan = gr.Chatbot(avatar_images=["human_icon.jpeg", "ai_icon.png"], height="auto", container=False, label="Common Plan History", likeable=False, show_share_button=False, show_label=True, elem_classes="chat-common", layout="bubble")
                        Chatbot_Common_Plan.change(lambda x: x, inputs=[], outputs=[], scroll_to_output=False, js=js_code)
                    with gr.Column():
                        gr.Markdown("## 📢 Ground Module Process History", elem_classes="markdown-text")
                        Chatbot_Common_Ground = gr.Chatbot(avatar_images=["human_icon.jpeg", "ai_icon.png"], height="auto", container=False, label="Common Ground History", likeable=False, show_share_button=False, show_label=True, elem_classes="chat-common", layout="bubble")
                        Chatbot_Common_Ground.change(lambda x: x, inputs=[], outputs=[], scroll_to_output=False, js=js_code)
                
                # Display chat history when button is clicked
                btn_show_history.click(fn=display_chat_history, inputs=[selected_models], outputs=[task, Chatbot_Common_Plan, Chatbot_Common_Ground, image])

            with gr.TabItem("📮 About Us", elem_id="od-benchmark-tab-table", id=3):
                gr.Markdown(ABOUT_MD, elem_classes="markdown-text")
        gr.Markdown(f"Last updated on **{LAST_UPDATED}**", 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("--result_file", help="Path to results table", default="data_dir/elo_ranks.all.jsonl")
    parser.add_argument("--length_balation_file", help="Path to results table", default="data_dir/elo_ranks.length_ablation.all.jsonl")
    parser.add_argument("--skip_empty_result_file", help="Path to results table", default="data_dir/elo_ranks.skip_empty.all.jsonl")
    parser.add_argument("--skip_empty_length_balation_file", help="Path to results table", default="data_dir/elo_ranks.skip_empty.length_ablation.all.jsonl")
    args = parser.parse_args()

    LAST_UPDATED = datetime.fromtimestamp(Path(args.result_file).stat().st_mtime, tz=timezone.utc).strftime("%Y-%m-%d %H:%M:%S")
 
    original_df = pd.read_json(args.result_file , lines=True)
    ablation_df = pd.read_json(args.length_balation_file, lines=True)
    skip_empty_original_df = pd.read_json(args.skip_empty_result_file , lines=True)
    skip_empty_ablation_df = pd.read_json(args.skip_empty_length_balation_file, lines=True)
   
    
    # available_models = sorted(list(set(list(original_df["model name "])))) 
    available_models = list(model_info.keys())
    # remove the rows where the model name is not in the available_models
    original_df = original_df[original_df["model name "].isin(available_models)]
    ablation_df = ablation_df[ablation_df["model name "].isin(available_models)]
    skip_empty_ablation_df = skip_empty_ablation_df[skip_empty_ablation_df["model name "].isin(available_models)]
    skip_empty_original_df = skip_empty_original_df[skip_empty_original_df["model name "].isin(available_models)]

    model_len_info = json.load(open("model_len_info.json", "r"))

    original_df = post_processing(original_df, model_len_info)
    ablation_df = post_processing(ablation_df, model_len_info)
    skip_empty_original_df = post_processing(skip_empty_original_df, model_len_info)
    skip_empty_ablation_df = post_processing(skip_empty_ablation_df, model_len_info)
    

    

    TYPES = ["markdown", "number"]

    demo = build_demo(TYPES)
    demo.launch(share=args.share, allowed_paths=["."], height=1000)